YT YouTube2Text

YouTube2Text → Transcripts → how to actually use ai for trading

how to actually use ai for trading — Transcript

by Moon Dev · 45,437 words · 7,507 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00I'm going to show you how to actually
0:02use AI to trade so not like everybody
0:06else I'm going to show you how to
0:08actually use artificial intelligence to
0:10trade and it's not the way you're
0:12thinking no it doesn't just predict
0:14price I'm sorry to break it to you right
0:16here right now but I'm going to show you
0:17the actual way how to use AI to trade
0:20now I've been at this for over 3,000
0:23hours today I'm going to show you
0:24everything step by step over the course
0:26of this short short video I'm going to
0:28hold your hand through the two most
0:31effective ways to how to actually use AI
0:34to trade by the end of this video you
0:37will
0:38know exactly step by step how to use
0:43AI there's two unique approaches that I
0:46show you through this video number
0:50one no I'm not having AI predict price
0:53no way the reason is because if
0:56everybody uses AI to predict price
1:00the price is going to change and the
1:02price is not going to be that predicted
1:03price anymore does that make sense let
1:06me say it
1:07again if everybody uses AI to predict
1:12the next minute price or hour price or
1:15day price of any
1:17asset then that price is not going to be
1:20the price anymore because people are
1:22going to buy and sell around it it's not
1:24like the weather when you predict hey
1:26tomorrow is going to be sunny well if
1:282,000 other people predict tomorrow is
1:31going to be sunny it doesn't matter it
1:33can still be
1:34sunny or if you predict there's going to
1:37be six car
1:40crashes yeah that's morbid I don't even
1:42want to talk about that if you predict
1:45that your your sales numbers in your
1:47Fortune 500 company is they're going to
1:50have 500 sales or something and then you
1:53have other 20 other data scientists
1:55predicting there's going to be 500 sales
1:57that's not going to change the sales but
1:59in
2:01trading if everybody's predicting
2:04price that's going to change the price
2:06so that prediction is not going to work
2:08that makes sense right so I'm sorry to
2:10burst the bubble right off here but I'm
2:12going to show you two legit ways to
2:14actually use AI to
2:16trade the first way to use AI to trade
2:19is the most obvious way dude we have
2:22Claude we have chat GPT we have luxie we
2:27have all of these different AI platforms
2:30that we can just ask questions now you
2:32do need to kind of know how to code in
2:34order to get the most out of them I
2:36think that they gave us way too much
2:38power with these AI platforms to be
2:41real but the thing is nobody knows
2:43because nobody takes the time to learn
2:45how to code so they don't see how much
2:47power we actually
2:49have I literally sit here I build
2:51trading systems that used to take months
2:54if not years by hand and multiple team
2:56members so that's the first way we're
2:58going to use AI we're going to use llms
3:00those are large language
3:02models to help me take my ideas out of
3:06my brain and put them down in code so
3:08then I can go ahead and test them I
3:10follow this system all day long this is
3:12my simplified system to algo
3:15trading so algo trading is the process
3:18of automating your
3:20trading and I start with researching and
3:24then back testing so llms AI can help
3:27with researching and back testing
3:29because it can take an idea and expand
3:31on it it knows everything I only know
3:33some things I don't know
3:36much and then it can help me code out
3:38these back tests to see if that research
3:41those strategies I came up with if they
3:44actually work in the past so llms are
3:47the first way I'm going to use AI but
3:48the second way is rest in peace to his
3:51soul his family everybody that loved him
3:53I loved him but not like anybody loved
3:56his family member you know how that goes
3:58but rest in peace the king Jim Simons
4:00Jim Simons if you want to Google him he
4:03is the best algorithmic Trader ever to
4:06touch this Earth and we've been studying
4:09him and you're going to see a lot of it
4:11here in this video but one of his
4:14favorite models before I get there he
4:18said he did not he let the he he made
4:21all strategies based off of data so if
4:24you're making strateg trading strategies
4:26based off of data that kind of infers
4:28that you're using machine leared because
4:31if you're starting at a Zer Point origin
4:33with just a bunch of open high low close
4:35volume data and some math well that's
4:38telling me that he said he starts with
4:40data that tells me he's doing machine
4:42learning we watched a ton of videos and
4:46he finally said it he really likes the
4:49hidden Markoff model the hidden Markoff
4:52model is a model a machine learning
4:56model that helps predict different
4:58states now I know this might be
5:01confusing but I go over it so many times
5:03throughout this video so it's going to
5:05get less confusing it was confusing to
5:07me too but the hidden Markoff model
5:10which I refer to over and over again as
5:13hmm is a machine learning model that
5:16helps predict hidden States now what
5:19does that mean
5:21mondev well if you go over here to some
5:24data here you can see there's like a a
5:27bull market State a regime so State and
5:31regime I'm going to use them
5:32interchangeably a lot in this training
5:35you can see this is like a sideways a
5:37consolidation state or regime this is
5:40like a bullish regime this right here is
5:42like a bearish
5:44regime so Jim Simons repeatedly said he
5:48likes the hidden Markoff model which is
5:50a machine learning model and machine
5:52learning is
5:55AI so not only are we going to use llms
5:59probably the the newest form of
6:02AI at least that we have access to in a
6:05userfriendly way but we're also going to
6:08use the hidden Markoff model which is a
6:11machine learning model that Jim Simons
6:14the goat used by the end of this you're
6:17going to have a machine learning model
6:21that predicts different regimes and it
6:24actually back tests dang I should pull
6:27up the back tests right now so you can
6:30see it let's see if I can pull it up
6:33really quickly if I can't then you're
6:35going to have to watch the whole thing
6:37dude I just wanted to CH you show you
6:40real quick so you can believe me I want
6:43you to stick around and watch this
6:45entire video because it's a journey
6:48dude it's a journey and you're going to
6:51learn so much and by the end of it
6:53you're going to have every single thing
6:55that I know about hidden markof models
7:00about using llms and AI in order to code
7:04for you help you flesh out your ideas
7:06you can see here these are a bunch of
7:08back tests here that we outputed and for
7:11example this first one has a return of
7:1452% while Buy and Hold is
7:1835% this is just the start we're just
7:20getting started with this stuff dude
7:23it's
7:25wild and the thing I like most about
7:28this is some of of these returns that we
7:32see they're only
7:34holding 11% of the time so that's one of
7:38the best returns that I can remember I
7:40didn't show you it right there I didn't
7:41have it prepared I'm sorry let's go
7:43ahead and just dive in and show you step
7:44by step by the end of this you're going
7:46to be able to know everything I know
7:48about how to actually use AI in trading
7:52and I show you everything step by step
7:55as always I go live every single day and
7:58I show you everything every single piece
8:00of code because I believe code is a
8:02great
8:03equalizer I believe if you know how to
8:05code you can pull yourself out of any
8:08position because you can build for the
8:09rest of your life and like I said I
8:11think they gave us too much power here
8:12with this this AI these llms I think if
8:16the the higher ups they knew about how
8:18much power coders have now with AI they
8:21would cut it
8:22off that's why I'm going so hard every
8:25single day because I don't think we'll
8:27have access to this forever I hope I'm
8:29hopeful we
8:31do cuz the can's already out the the
8:34worm's already out of the can whatever
8:36the saying is I hope we do but I'm going
8:39to show you everything step by step here
8:40right now let's get into it stick around
8:42for the whole thing Jim Simons love the
8:45hidden markof model so I wanted to
8:47investigate a little bit more got all
8:49the code written out so it looks like
8:51we're just importing pandas here and
8:54import numpy MP and then from hmm learn
8:57import hmm Google this B pip install hmm
9:03learn then map plot li. pyplot as PLP
9:07PLT PLT SK learn. preprocessing okay
9:10we're starting the hmm analysis here I'm
9:13using like uh what year is it it's 2024
9:17so six years of BTC hourly data it's
9:20data 51,9 or 5
9:2551,9 122 rows of
9:28data um
9:29I guess I could get more I could
9:31absolutely get more absolutely
9:33absolutely and different time frames but
9:36I just want to get this set up and
9:37working and then worry about all that
9:40stuff
9:43later but
9:45essentially loading in the data okay
9:48we're naming the files open high low
9:51close volume pd. read file path names I
9:55don't think that's right but let's go
9:57ahead and look at it daytime yeah I
9:59don't think that's right but we're just
10:01going to pretend like it is for now and
10:03no let's fix it let's fix it date time
10:06date time open high low close pd. read
10:10file names what if we just dro the read
10:14yeah yeah what if we did
10:16that okay okay and then just say print
10:20data let's start there print
10:23DF doad fine time. sleep oh we don't
10:27even have time in here
10:30AI sometimes
10:32huh load pre-processing data okay let's
10:35go down to the bottom
10:37here and it says training start
10:40executing uh starting main execution
10:43okay and then it starts this okay we'll
10:46start there so we'll just start there
10:48and then go step by step because then
10:50you get to see all the code I get to see
10:52all the code and then we all get all the
10:55code you
10:57know you see the code I see the code we
11:00see the code and then we can do cool
11:03things with the code maybe but I did
11:06watch that video the other day that you
11:08know compared the the hidden Markoff
11:11model hidden I'm going to keep writing
11:12that out because it's such a nerd word
11:14hidden markof
11:16model uh
11:18was the was better less than the RNN the
11:24current neural
11:27network it was less than let but it's
11:30only on one day only on one day and one
11:34person's features really because this is
11:37just feature engineering right that's
11:39what that one dude said at least Andrew
11:42ning I'm so sorry to Butch your name if
11:46that's true you I love you so so much yo
11:50they really they
11:52really Andrew really has all of these
11:55things let's go ahead and watch a couple
11:56of
11:57them not right now
12:01but
12:03maybe this is look at that deep learning
12:06AI stop
12:16it I feel like he had like a some class
12:18at some
12:21school oh this right
12:24here stanf or
12:28something
12:29engineering this is
12:32uh I watched a good amount of
12:38this dude I think we watched this whole
12:51thing okay so the reason I guy here is
12:53because he has said something about like
12:56it's just feature engineering it's like
12:58that's where the edge is and machine
12:59learning is feature engineering it's
13:01like coming up with the ideas what you
13:02going put what data you can put in what
13:03are the uh yeah that's it that's all I
13:07got I learned I learned machine learning
13:10before learning how to code it's crazy
13:11that's how I learn how to code so
13:13whatever it is what it is weird path but
13:16I'm going be able to pick this up easy
13:18dude not easy
13:21but it's like a refresher that's it
13:25that's it dude so lecture 19 we might go
13:28up in there any anyways let's go ahead
13:29and print out this data
13:33here boom got it unnamed
13:39six yeah my data is janky dude I forgot
13:43about that this is good this is a good
13:46practice how do I do that ah how do I
13:49select all the rows no all of these
13:54there's a way to do that um let's go ask
13:57AI
14:04okay so what I'm going to say here oh my
14:06God I love this what am I doing this is
14:08amazing I can't believe it I can't
14:10believe the opportunity we have here
14:12with AI we can ask anything
14:15so my data above is super
14:19janky it has a comma at the end of the
14:26column names indicating
14:30a new
14:31column how do I I know I could just drop
14:35it probably I can I can drop I know I
14:37can drop that column okay I got that but
14:40I've seen people do this little thing
14:42where they select the end column and it
14:46it's
14:47like a keyboard shortcut and it selects
14:50you have a selector on every line yeah
14:53that's it all right how how do I make it
14:56so on the CSV I I use that keyboard
15:02shortcut to
15:05um have a
15:08selector door blinker I don't know what
15:11these are called blinker on each line
15:15and then
15:16delete the last
15:19column I know I can
15:22drop with pandas but I'm curious about
15:27that keyboard
15:30shortcut and how to do the same thing on
15:36on multiple multiple multiple
15:41lines okay you know what I'm talking
15:44about though right you know what I'm
15:45talking about I know you know what I'm
15:47talking about I see you're looking for a
15:49way to efficiently edit multiple lines
15:50yes there we
15:53go and your text editor remove the
15:55trailing commas this is a common task
15:57and many text editors have features to
15:59help us the T teni the ten the teni
16:06uh the technique you're referring to is
16:09often called multic cursor editing or
16:11column selector here's how to do it in
16:13some popular so press alt shift and I
16:18that's it that's all I need to know now
16:20I'm I'm off
16:22alt option
16:26maybe no control
16:34no dang what's all on
16:38Mac
16:39controll
16:41shift
16:45I oh okay option shift to add
16:50cursor end of each line okay let's try
16:54that
16:56option shift
17:01I maybe this not an
17:08i no I mean option shift I does it have
17:11to be in order maybe the same
17:23time n dude ain't
17:27it visual Studio
17:31code I'm just going to drop it I'm just
17:33going to drop it I'm just going to drop
17:34it because it doesn't work for
17:37me blinker
17:39cursor bro you should throw that data
17:43into a DB like postgress or my
17:47[Music]
17:50skill my
17:54skill how
17:57come how come
18:00that is the
18:07question I'm just going to drop it so
18:09let's go back to it my bad my bad I
18:12thought I would be able to pick that
18:14pick that up real quickly and it
18:17couldn't so
18:21drop drop the last
18:25column thank you okay it should be gone
18:28now look how much easier it is to do
18:30things with
18:31code you can't tell me code does not
18:34give you advantage in life can't tell me
18:36that I'm sorry I've been I've been
18:39living I know I'm not the oldest old
18:41head out there but I've been living and
18:43this stuff is Magic
18:46dude drops a column calls coding magic
18:50let's see print creating date time range
18:53okay let's say this let's say print
18:56creating date time in index calculating
19:00returns and volatility I'm not going to
19:01PR everything that'll take forever
19:03calculating returns and volatility okay
19:04the returns is the percent change of the
19:06close Okay volatility is returned the
19:09rolling standard
19:11deviation of a rolling window dang this
19:14is some good stuff be volume change
19:17volume percent change okay dropping nonv
19:19values data process and let's just print
19:22out our DF let's start there make sure
19:25everything's looking good this is we're
19:28calling the main main execution down
19:29here this is the main execution we just
19:31went through the loading and pre-process
19:32data okay you saw it all
19:36dude print
19:38data not not I want all of it though
19:41whatever I'll take the I'll take that
19:44I'm not sleep five no way 555 deal
19:48remember that you remember the 555 deal
19:50at
19:53dominoes I was
19:56good I was good I think I'm a little
19:58hungry so these returns has to be a
20:01lowercase close this needs to be a
20:04lowercase volume who did this
20:07dude should have gone with industry
20:10standards no clothes in it okay well
20:13let's go look at it close looks like a
20:15close to me did I spell it
20:18incorrectly I have to make the columns
20:20yeah I got to make the columns okay
20:22that's crazy because there are columns
20:24here but
20:38let's see
20:40here let's see what line is this on 201
20:43108
20:45um
20:4627 here we go so returns it's having an
20:51a right there no it's not it's on
20:54returns this one right here so close do
20:58you have close
20:59CL print the columns then show me what
21:02the columns
21:04are yo it's so funny because there's no
21:07reason to
21:08code stop
21:11it I don't get it how did we get here
21:15because I would rather just write in
21:16English you know that's what I'm saying
21:18like I I know I know I need to print out
21:20the columns to check it out but it's
21:21like okay easier way to play the game
21:23dog easier way to play the game dog
21:27daytime open yo these are janky look at
21:31this there's a space in
21:38between that is
21:43crazy that is insane all right so I'm
21:46just going to go say close and then
21:48volume
21:49here I'm going to say rename the columns
21:53this is easy this is light work when you
21:55when you have this stuff rename rename
21:58the call
22:00SS two open high low close volume
22:04dude thank you D have. columns okay
22:07let's just confirm
22:09it ah no no no that ain't it that ain't
22:12it we got to have date
22:19time Daye
22:23time date time that is so janky come on
22:27dude come on you got got
22:29this there we go daytime now let's see
22:33what the columns are looking like okay
22:36perfect the columns look good now just
22:38getting things all
22:40formatted okay time. sleep no more sleep
22:42now let's go back come on bring it back
23:08okay so now we have the close
23:09calculating returns of volatility
23:11perfect perfect perfect creating the
23:13volume this should all work now I mean
23:16we'll see but let's run
23:18it okay great so we got the returns here
23:22we've got the volatility here and the
23:24volume
23:26change perfect
23:28all right let's keep it moving
23:30then I think it's important to
23:32understand this codebase because this is
23:34like just a just a start B just a
23:38start with
23:41path all right so training hmm model
23:45okay so let's go check out the training
23:47so we're passing in this data
23:49here passing in
23:53the the above data to Now train
23:59the train train
24:02the hmm
24:05model OKAY training hm model let's go to
24:10train definition get go to definition
24:12okay so print training hmm with n
24:20components the features are returns
24:23volatility and volume
24:27change x equals data of the features
24:32values it's got the standard scale so
24:34it's normalizing the features standard
24:36scaler and we're going to go through all
24:37this don't worry don't worry scalar
24:39equals standard scalar X scaled equals
24:42scalar so we have to fit and transform
24:44it and then we have to fit
24:49it fitting hmm model and then this is
24:53that pip andall that we made earlier
24:56hmm. goian hm m n components equal n
25:00components ciance type equals full n
25:03iterations random state so the random
25:05State 42 model that fit okay this is the
25:09actual training of it I
25:11believe hmm training complete okay I
25:14don't think this works I think we we get
25:16into an error here so first
25:23off let's explain it so we can kind of
25:26understand and help the AI out
25:31explain the
25:33above the above code with two different
25:39uh examples that a 12 year all can
25:48understand um yeah let's see what are
25:53components uh what are components
25:59um n
26:04iterations covariance
26:10types
26:14um standard
26:18scaler all right this is good fit
26:23transform this is good this is good
26:26okay so certainly I'll explain the code
26:29using two simple examples that a
26:3112-year-old might
26:33understand this secret mood detector
26:36imagine no I can't do this B the secret
26:40mood detector all right imagine you have
26:42a magical mood detector that can figure
26:45out if your friend is happy sad or
26:47neutral based on three things how much
26:49they smile how loudly they speak and how
26:52fast they walk the components in this
26:55example the components Nal 3 n component
26:58equals three are the three moods happy
27:02sad and
27:05neutral the features are smile amount
27:09speaking volume and walking
27:13speed the standard scaler is like
27:15adjusting your eyes to see these
27:17features
27:19clearly no matter if your friend is a
27:23quiet Walker or a loud
27:26Runner fit transform is like calibrating
27:29your mood detector with all the
27:31information you've
27:32gathered and it
27:37equals means the detector will practice
27:41guessing moods a 100 times to get better
27:43at it yo this is actually pretty good
27:45covariance type equals full means the
27:49detector considers how all features
27:52might be
27:54related like maybe people walk faster
27:57when they speak
28:02loudly the gazum hmm is your magical
28:05mood detector that learns patterns to
28:08guess your friend's mood even when you
28:11can't ask them directly and remember
28:12we're trying to guess the market regimes
28:14maybe you don't remember maybe you
28:16weren't here I don't know come on bro
28:19keep it keep it up keep going that's it
28:21come on we going example two the weather
28:24prediction game let's say you're playing
28:27a weather pred addiction game you can't
28:29see outside but you have three
28:32Clues temperature humidity and wind
28:36speed you're trying to guess if it's
28:39sunny rainy or
28:41cloudy in this scenario the components n
28:44equals 3 n components equals 3 are the
28:47three types Sunny rainy and
28:51cloudy so that could be
28:54like trending
28:57consolidation downward consolidation
28:59upward I mean what Jim said there's like
29:01nine I'd be curious but those components
29:04could
29:05be could be uh those the different reg
29:10regimes
29:12regimes the features are temperature
29:14humidity and wind speed the standard
29:15scaler is like converting all
29:17measurements to a scale from 1 to 10 so
29:20it's easier to compare them fit ter
29:23transform is like studying lots of
29:25weather reports to understand what
29:26combinations usually mean
29:29n it equals 100 means you'll practice
29:33guessing the weather 100 times to
29:34improve your
29:36skills covariance type equals full means
29:39you consider how temperature humidity
29:42and wind might affect each
29:45other the gajian hmm is your brain
29:48trying to learn patterns to guess the
29:51weather
29:52correctly even when you can't look
29:55outside give me two examples
30:00me two
30:02examples of trading
30:07alos and
30:09boss for crypto
30:12um I mean I understand it but again this
30:15was hard for me to get through my thick
30:19skull oh I didn't want actually code
30:21examples
30:28I'm just going to move on let's see if
30:30this
30:33works there's plenty of videos on this
30:36people much much smarter than
30:38I okay so look at this look at that
30:42that's dope actually because yesterday
30:45this didn't work but Mund went up in
30:47there and raw doged it a little bit and
30:50fixed
30:51it if you remember the end of yesterday
30:54it was like oh well we did good but
31:02um there's an error is there still an
31:05error I don't want to start acting like
31:07I figured it out or anything oh I got to
31:09sleep there okay okay stop it I was
31:12hyping myself up real quick all right
31:14get off me get off
31:19me oh my goodness
31:22so explain what the outcome
31:29this should send back our return the
31:31model does it actually did it say
31:33training complete I don't think it did I
31:36think it was training right there
31:40yeah States predicted let's go ahead and
31:44say where does it say States
31:45predicted okay predict States dude we
31:49didn't go over this
31:54one hmm training complete okay so we
31:59didn't go over this part predicting
32:01States there's only three
32:15states analyzing States okay got you got
32:18you there's a pause there that's our
32:20pause that's fine but we didn't go over
32:22this part so
32:29I'm going to say explain this one
32:31explain this
32:33code explain this
32:36code with two examples not examples with
32:41uh the theme being trading alos and
32:47trading Bots
32:58okay so certainly let's explain predict
32:59States function so we're looking at this
33:01function here just so you can see
33:10it in the context of trading algorithms
33:12and Bot for cryptocurrency markets
33:13imagine you're running a sophisticated
33:15crypto trading bot that doesn't just
33:17look at price but tries to understand
33:19the overall mood or regime of the market
33:21this function is like the Bots Market
33:25psychologist trying to figure out what's
33:27date the market is in at any given time
33:32here's how it works features equals the
33:35returns the volatility and uh volume
33:38change and we have all that remember I
33:39printed it
33:43anyways your Bot is looking at three key
33:47aspects of the market returns how much
33:50the price has
33:52changed volatility how wildly the price
33:54is
33:56swinging volume change change how
33:58trading activity is changing this is
34:01like a Trader watching not just the
34:03price but also how nervous the market
34:05seems and how many people are
34:09trading x equals
34:12data x equals D this line here x equals
34:16data passing in features or in the
34:18features section values debot collects
34:22these three prices or pieces of
34:24information for each time period it's
34:27looking at
34:29X
34:31scaled this is like normalizing the data
34:34Maybe returns
34:35are usually tiny like 0.1% but volume
34:39change can be 50% the scaler makes sure
34:43the bot doesn't overreact to naturally
34:45large
34:48numbers that's cool but finally
34:51clicks this is where the magic happens
34:54States equals model.
34:56predict EX
34:59date so this is where this is where the
35:01magic happens the bot uses its trained
35:04model like an experienced Trader in int
35:07Traders intuition
35:10intuition to guess what state or mood
35:12the market is in based on the returns
35:15volatility and volume changes it
35:18seen print States predicted unique
35:22states the bot announces how many
35:25different states it found
35:28for example it might say it found three
35:32states which could be something like
35:35state zero calm Market low returns low
35:38volatility normal volume state one bull
35:42market High positive returns medium
35:46volatility and high volume Panic Market
35:49High negative returns High volatility
35:53and very high volume
35:58in the context of a trading bot this
36:00function would be called
36:02regularly maybe every hour or day to
36:05reassess the market State the bot would
36:09use these State predictions to make
36:12trading
36:13decisions for example in state zero
36:16which is a calm Market it may might make
36:19small careful trades
36:37what is Dar I don't know in state one
36:40bull market it might make larger buy
36:42orders in state two a panic Market it
36:45might sell to protect assets or make
36:47short trades the key idea here is that
36:50the bot is trying to adapt its strategy
36:53based on what it perceives as the market
36:56current market i s just like a human
36:59Trader would it's not just blindly
37:01following a single strategy but trying
37:03to understand the Market's mood and
37:06adjust
37:09accordingly this approach can be
37:11particularly useful in cryptocurrency
37:13markets which can rapidly switch between
37:15periods of calm explosive growth and
37:19sharp
37:20declines by detecting these states a
37:23trading bot can potentially make smarter
37:26decisions when to be aggressive when to
37:28be cautious and when to stay out of the
37:30market
38:21just move on I mean I understand that to
38:23you I hope so I mean if you don't that's
38:25fine though because it's this is like
38:28super deep stuff you know it's deep it's
38:33deep it's deep let's look at analyze
38:36stats
38:39States all right so print analyzing
38:41States DF do analysis is a data. copy
38:45and then analyze State equals state for
38:49state in range of model. n components so
38:53it grabs the components analyzing state
38:58State data equals
39:00DF where State equals
39:03State print State returns volatility and
39:07Dot
39:08describe number of periods in
39:12state
39:16okay let's just print it oh there's
39:19nothing to
39:20print plot results let's check out plot
39:26results print plotting results so we're
39:28getting subplots PLT do subplots so our
39:32good old PLT I've never understood why
39:34it's called PLT let's try to figure that
39:35out right now Matt plot lib as PLT what
39:40are you talking about my dude I don't
39:42know I don't know plot maybe yeah maybe
39:45because plot okay got it figured it out
39:48plot plot without the oh
39:50[Music]
39:52the ax1 plot index data
39:57okay we got the close now I'm glad I
39:59switched it to close because it's just
40:01all up in here bitcoin price in hmm
40:04States okay for state and range of
40:06model. end components mask equals what
40:09is a
40:22mask what is a mask
40:27great question in the context of data
40:29analysis and visualization a mask is a
40:31Boolean true or false array that's used
40:34to select or highlight specific parts of
40:36your data let's break this down in a way
40:39that's easy to understand imagine you're
40:41looking at a long series of
40:42cryptocurrency prices and you want to
40:45color code different periods based on
40:48the market State like bull market bare
40:50Market or sideways
40:53Market The Mask helps you do exactly
40:56that here here how it works in the code
40:58for state and range this Loop goes
41:01through each state the model has
41:04identified let's say there are three
41:12states 0 1 and
41:17two mask equals State equals States this
41:20is creates a Boolean array mask where
41:24it's true for every time point where the
41:28Mark is in the current
41:35state okay got
41:39it so it's just like true or
41:41false let's use a simple example to
41:45illustrate suppose you have 10 days of
41:47data and the states array looks like
41:50this States equal 00
41:520111
41:5422 011
41:58this Loop goes through each
42:01state the model has identified let's say
42:04there are three states 0 1 and two mask
42:07equals States equals State this creates
42:09a Boolean array
42:11mask where it's true
42:15okay when creating mask for each state
42:18You' get true true false false
42:22false false false false okay I got it so
42:26it's for each state
42:28so you get two it's for you that's where
42:30I was a little confused like where's
42:31this two come from but it could be a
42:33seven it could be a nine it could be
42:34whatever for State 9 is it true or false
42:38you can see for State one here it's not
42:41going to be true until 2 three 4 or
42:44whatever that is 2 three 4 all right
42:46figured it out understand it let's keep
42:49it moving B in the context of a trading
42:51B algorithm these masks could be used to
42:53back test strategies for example you
42:55might want to see how particular trading
42:58strategy performs only during periods
42:59identifies State
43:13one
43:20duh okay okay okay you could use these
43:24Mass to calculate State specific metrics
43:27like what's the average daily return
43:30since when we're in state
43:322 in live trading identifying the
43:35current state could trigger different
43:37trading rules or risk management
43:40strategies yeah I feel like I should I
43:43don't know why I say explain it to me
43:45like I'm 12 just explain it to me for
43:48outo trading
43:50always here on out all right you heard
43:53it here first Mev live
43:57Live on
44:01YouTube live on YouTube every single day
44:05dude I don't know why I just like to
44:08show it I like to show this stuff
44:10opposed to just doing it by myself I'm
44:12sorry it's just like kind of like deep
44:15it's deep you know it's deep so like I'm
44:18not going to be able to comprehend all
44:19of it that's it that's it I can't
44:22comprehend all
44:23this you got to see it bro
44:27you you you you have a hack at this
44:30stuff dog I'm G show you everything
44:32every day every day okay ax. Legend so
44:36we're just building a legend here for
44:38the uh ax one and two so it looks like
44:41we're getting two plots I wonder if it
44:43will work let's go ahead and just run it
44:48up plot plot plot plot time do sleep I
44:52don't want to see the err that's why I
44:53keep putting this these time do sleeps I
44:55want to understand every single line of
44:56code
44:58you know why not okay so plotting
45:02results yo pzy plots pzy plots pzy plots
45:07a lot of
45:09thoughts get your head out the gutter be
45:12in your mind thoughts in your
45:16mind all right
45:18so I don't know
45:21but you know what we can do from here
45:25hey guess what guess what
45:27guess what guess what we can do from
45:30here dude we can do anything this is
45:33crazy this is crazy AI is nuts I can't
45:36believe it I can't believe what I'm
45:38about how I'm about to chop this
45:39up chop it up explain to me like
45:45I'm however old I
45:48am okay and in the
45:51context of trading Bots and trading
45:55algorithms
45:58we are building out a uh hidden
46:02markof model and Below attached
46:08attached no no no attached is the
46:12plotted
46:14stats I don't know stats or States
46:18now please explain it to me please
46:22explain with
46:24examples okay let's see it not code
46:26examples though don't give me any code
46:30hey not right now no thank you certainly
46:33let's break down what this hidden
46:35Markoff model plot is showing us in the
46:37context of trading Bots and algorithms
46:39for BTC the hidden Markoff model
46:42overview in this context the hmm is
46:45trying to identify different hidden
46:47states of the BTC Market these states
46:50are directly observable but are they
46:53aren't these states aren't directly
46:55observable but are infirm inferred from
46:59the visible data okay let's look at this
47:02data too let's just look at it while
47:05while I'm reading it because
47:07then that's how you do this stuff
47:13[Music]
47:15dude all right
47:19so these states aren't directly
47:21observable but are infirmed inferred
47:23from the visible data price returns
47:26volatility Etc the three states the
47:28model has identified three distinct
47:31States state zero 1 and two represented
47:35by different colors in the top of the
47:38chart so you can see three states zero
47:41one and two zero is blue one is yellow
47:45oh I'm sorry that's like orange or peach
47:47peach and two is
47:50green all right let's get it
47:57state zero this appears to be the most
47:59common state it might represent normal
48:02or Stable Market condition State one
48:06orange this state appears less
48:08frequently and often coincides with
48:10price increases it might represent
48:13bullish or a volatile uptrend
48:27this state is least common it repres it
48:29might represent a bearish or highly
48:31volatile
48:33State price chart
48:36top I mean let's start doing to-dos here
48:40to-do too too
48:43too I'mma keep it in the ideas I'mma
48:46keep it in the
48:47ideas um
48:59put um all types of States look into I
49:06can't get my words together look into
49:09look into Jim's
49:13favorite
49:15favorite yeah gy regimes regimes and
49:20make those the states that are
49:24predicted I guess
49:27guess what that's what I'm I'm trying to
49:28say okay so whatever this St is least
49:32common okay highly volatile I haven't I
49:33don't see any greens on here to be
49:35honest do you make it big
49:41bigger I don't see any
49:46greens that's
49:49okay
49:52um the Blue Line represents btc's price
49:55overtime I think that was
50:10obvious the colored backgrounds show
50:12which state the model believes the
50:15market is in at each point
50:26returns chart which is at the
50:32bottom this shows the percentage returns
50:35of BTC spikes indicate large price
50:40movements up or down so
50:45this is
50:48co and that coincides right with the
50:50price so this is lined up which is nice
51:04that hourly though this is hourly since
51:082018 on
51:14BTC I want to put Liquidations in there
51:18I don't
51:20know all the data that we
51:22got anything throw it all throw it all
51:26at it
51:28you got ideas throw it in the chat
51:35777 every single day you're up inside my
51:39brain
51:40b it's kind of weird you you up been
51:42here a lot
51:44honestly so just share your your 777s in
51:48the chat if you got a good
51:51idea share that with your brethren meet
51:54somebody in the Discord you don't have
51:55to share with me dude I just 1 plus one
51:59I just want 1 plus one to equal three
52:01like that one dude
52:04said all right price
52:08chart trade B applications example one
52:11state based strategy a Trad B could be
52:14used could use the states to adjust the
52:17strategy use a conservative strategy
52:19during light blue maybe scalping small
52:23profits State one orange Implement more
52:25aggressive buying strategy anticipating
52:27price
52:32increases in state
52:34two green either stay out of the market
52:37or Implement a shorting
52:41strategy example two Trend
52:43following notice how State one orange
52:46often appears during uptrends
53:03aach could use this to enter long
53:05positions when transitioning from 0o to
53:08one 0 to
53:12one exit or reduce positions when
53:15transitioning out of state
53:18one example three volatility trading the
53:21returns chart shows periods of higher
53:24volatility a bot could increase position
53:27sizes during state zero when volatility
53:30is
53:34lower reduce risk and tighten stop
53:36losses during States one and two when
53:39volatility might be
53:41higher example for mean reversion after
53:43extended periods in state one or two so
53:47one or
53:52two okay I mean I I think this data just
53:57needs to be I don't know I think we just
54:00need to keep going through this we're
54:02not even done we're not even done
54:04yet
54:06so risk management the rapid State
54:08changes in some periods like early 2022
54:11indicate a highly uncertain Market a
54:14trading algorithm might reduce its
54:16overall exposure during such times
54:19Market regime detection the long
54:22stretches of state zero in 201819
54:26might indicate a bare Market or
54:28accumulation phase a sophisticated
54:30algorithm could adjust its long-term
54:33strategy based on these Extended State
54:35periods in essence this hmm is providing
54:38a framework for a trading bot to
54:42understand the current market context
54:44Beyond just price it's like giving the
54:47bot a sense of the Market's mood
54:50allowing it to adapt its strategies to
54:52different Market regimes this can be
54:54particularly valuable in the volatile
54:58Andy cyclical nature of the
55:00cryptocurrency
55:05market okay so let's just keep it moving
55:07then let's see how much else like I know
55:09we got something else here so how much
55:12else we
55:13got so printing the transition Matrix
55:16let's go ahead and see that man I don't
55:19know if it's going to let us because
55:21we'll see let's just let's just see it
55:23let's just see
55:24it feel like it gets stuck on the the
55:27output of the the other thing you know
55:30plotting results yeah it's going to stop
55:32there showing plot okay I can close the
55:35plot now and then it should move on
55:37there we go transition
55:40Matrix
55:43okay I want to save this plot though
55:46analyze the stats predict the stats plot
55:48the results let's go save it dude let's
55:52save it to current directory tight
55:54layout
55:57pt. show save the save the plot to my uh
56:04data folder B in this directory all
56:08right that's it data dot okay but
56:16as as a plot B not as not as the data
56:21that's looking weird that's looking like
56:22the other
56:24thing okay let's go ahead and make a
56:27data folder here look at this look at
56:29this just just keep swimming be just
56:31keep swimming this is fun I like it if
56:35Jim liked it I like it all
56:39right get off me B get off me let's go
56:42ahead and check it out I don't even need
56:44to show it
56:45anymore I
56:48could it's plotting okay perfect perfect
56:52perfect it's still plotting but now I'm
56:54going to close the plot
56:57and then I see transition Matrix okay so
56:59I
57:01might go ahead and
57:06say n I'm not going to do any custom
57:08stuff there right now not the point
57:10transition Matrix okay let's go ahead
57:13and print out the transition
57:15Matrix let's get the stats of
57:21it oh snap
57:26okay yeah we've always got that
57:29though transition Matrix
57:32okay I'm going to copy that in a second
57:35I want to First go check out the
57:38transition Matrix walk ourselves through
57:40it see where we're at say see where
57:42we're at oh model. Trans okay there's no
57:46code there there's no code there dude
57:49there's not one piece of code there I
57:51mean mod model.
57:53transmat so I think that's just how you
57:55call the transation transition Matrix
57:58which is fine that's fine with me
58:01but mean State variance I just want to
58:04explain that explain
58:08this uh for my hmm can you you explain
58:15this
58:17transition
58:22Matrix what dude that's it stop it
58:36explain
58:37this transition
58:41Matrix
58:43please transition Matrix where are
58:48we I don't know I don't know anymore
59:05okay so certainly let's break down this
59:09transition Matrix in the context of your
59:11hidden Markoff
59:12model for BTC trading the transition
59:15Matrix shows the probabilities of moving
59:18from one state to another in the next
59:20time
59:22stamp in your case you have three states
59:26let's call them 0 1 and two The Matrix
59:29is read read as from state to
59:33state from
59:36state to
59:39state so transition from state
59:45zero you can see it's the
59:480.92 wait wait wait wait wait wait how
59:50are you going to change up on me oh it's
59:53negative 1 okay got it got it there's
59:55that to the 10th power thing up there so
59:58it's negative one so
1:00:0392078 okay and point blah blah blah blah
1:00:06blah
1:00:07254 92% chance of staying in state
1:00:13zero 7% chance of moving to state
1:00:19two and very small chance of moving to
1:00:22state two sorry one and then two
1:00:26transitions from State
1:00:32one there's
1:00:35a 24% chance of M moving to state
1:00:42zero and a 75% chance of staying in
1:00:46State One
1:00:58essentially 0%
1:01:00chance of moving the state
1:01:03to transition from State
1:01:082 100% chance of moving into state zero
1:01:13so that means there is no state
1:01:15two
1:01:18right interpretation for trading state
1:01:20zero likely normal or Baseline State
1:01:23very stable tends to process
1:01:26occasionally it transitions to State one
1:01:28rarely goes to state
1:01:30two State one possibly a bullish or
1:01:33trending State moderately stable often
1:01:37persists sometimes reverts to state
1:01:46zero sorry to wake you bro sometimes
1:01:50revers to state zero almost never goes
1:01:53directly to state two state two possibly
1:01:55a volatile or shock State extremely
1:01:59unstable never persists always
1:02:02transitions back to state zero
1:02:04immediately trading
1:02:06implications state zero could represent
1:02:08represent a calm Market a trading bot
1:02:11might use Conservative Strategies here
1:02:14State one seems to represent a
1:02:16persistent Trend a bot might use Trend
1:02:20following strategies when in this
1:02:23state state two appears to be a very
1:02:27rare instantly selloff correcting
1:02:30state it might represent sudden Market
1:02:34shocks or extreme events a bot should be
1:02:38prepared for quick reversals after
1:02:40detecting the
1:02:44state the Rarity and instant transition
1:02:47of State 2 back to state zero suggests
1:02:49it might be capturing outlier events in
1:02:52your data you might want to investigate
1:02:56what market conditions correspond to
1:02:57state two and consider if it's providing
1:03:01valuable information for your trading
1:03:02strategy it's not printing means and
1:03:05variances co-variances of each state
1:03:08okay let's see it
1:03:10dude yo I wonder if this is popping
1:03:13out sure is sure
1:03:18is cool cool cool we know how to code
1:03:22just a little bit just a little we know
1:03:24a little bit of code
1:03:26then we got AI my
1:03:28God come on let's
1:03:31go let's do it
1:03:34dude anything that's in your brain it's
1:03:36already been learned before dude stop
1:03:40it co-variance yes give me all that walk
1:03:44me through all this
1:03:47please print print print
1:03:52okay walk me through all of this now
1:03:58it's the last part of the prints of my
1:04:03code my code your code our
1:04:07code my code your code our code our code
1:04:44so let's look at it here let's close
1:04:47this up a little
1:04:49bit tieing up tieen up close that yes
1:04:53okay here we go we should get it here
1:04:56soon where you at let's see it why you
1:04:59moving so
1:05:01slow she's
1:05:03tired she's
1:05:07tired all
1:05:09right certainly let's walk through this
1:05:11output in the context of a hidden markof
1:05:13model for Bitcoin trading these
1:05:16statistics describe the characteristics
1:05:18of each state identified by your model
1:05:21for each state we have mean the average
1:05:24values of the feature in this state
1:05:28covariance how the features relate to
1:05:30each other in this
1:05:32state the features appear to be in this
1:05:34order returns volatility and volume
1:05:36change state zero mean
1:05:40okay slightly positive returns below
1:05:44average volatility in state zero and
1:05:47slightly decreased
1:05:51volume covariance shows low correlation
1:05:55between features interpretation this
1:05:58could be a calm or sideways Market State
1:06:01returns are small but positive
1:06:03volatility is low and volume is slightly
1:06:07decreasing State
1:06:12one the
1:06:15mean slightly negative
1:06:18returns above average
1:06:21volatility increased volume
1:06:27covariance shows stronger
1:06:36correlations especially between
1:06:38volatility and
1:06:39volume interpretation this might
1:06:41represent a volatile or uncertain Market
1:06:44State returns are slightly negative but
1:06:46volatility and volume are both up
1:06:50significantly state two strongly
1:06:53negative returns
1:06:56decrease
1:07:00VA extremely high volume increase
1:07:03covariance all values are
1:07:0701 which is unusual and might indicate a
1:07:11computational artifact or a very rare
1:07:14State interpretation this could
1:07:16represent an extreme event or Market
1:07:19shock returns are very negative
1:07:22volatility surprisingly drops and
1:07:24there's an enormous spike in
1:07:26volume trading strategy implementations
1:07:29implementations
1:07:31implementations in state zero use
1:07:33Conservative Strategies look for small
1:07:35quick profits be prepared for potential
1:07:38transitions to more volatile States in
1:07:41state one Implement risk management
1:07:43strategies due to high volatility look
1:07:46for short-term trading opportunities in
1:07:48both directions monitor for potential
1:07:51Trend
1:07:52formations in state to be extremely
1:07:55cautious
1:07:56this state might represent rare extreme
1:07:59events like Market
1:08:01crashes or major news
1:08:05events consider implementing stop losses
1:08:08or hedging strategies be prepared for
1:08:12quick reversals okay I want to see that
1:08:16though where it how can I see where they
1:08:19because I can't see it on the chart
1:08:21maybe we can maybe we just need to make
1:08:22it bigger let's make it bigger can you
1:08:24see this chart
1:08:27can you see
1:08:28it I hope you
1:08:30can I don't see any green up in
1:08:35here I see no
1:08:38green all right green green green green
1:08:43Green where you at Green I don't see any
1:08:45green maybe I'm color
1:08:47blind maybe I'm color blind but that's
1:08:50okay so where do we go from here I feel
1:08:53like these states need to be chopped up
1:08:54a little better
1:08:56so um below attached is my code attached
1:09:01is is my code I know that
1:09:06um Jim
1:09:09Simons said there were like eight plus
1:09:16regimes should it we have a
1:09:21regime uh state
1:09:25for at
1:09:27least six
1:09:31then I can think of a
1:09:35few please add them in and walk me
1:09:40through the changes okay so let's say
1:09:45bullish
1:09:48trending bearish
1:09:52trending uh sideways
1:09:57consolidation
1:09:59upward
1:10:01consolidation
1:10:03idation
1:10:06downward
1:10:08consolidation
1:10:11uh
1:10:14consolidation uh
1:10:17downward
1:10:20capitulation and
1:10:24upward
1:10:27capitulation
1:10:29right you got more dude 1 2 3 4 5
1:10:3467 so you got bullish trending there's
1:10:36not that many things in in the market
1:10:39right I don't know I don't know dude I
1:10:42don't know anything I'm just here I'm
1:10:44just here with you my dude me and you
1:10:47every day around this time or another
1:10:50time I don't know just when you see me
1:10:52dude come on get up on here all right
1:10:57so there you got like
1:11:01bullish you got
1:11:05bearish uh you got like sideways upward
1:11:09consolidation but then in in here it's
1:11:11like downward but then there's a
1:11:12capitulation downward there's a
1:11:15capitulation
1:11:17upward I guess I need to describe these
1:11:20areas yeah yeah let's do that okay I'll
1:11:23describe it to
1:11:25I'm not going to describe it um can use
1:11:28volume I wish I had the
1:11:30[Music]
1:11:32uh I do
1:11:35dude I was going to say I wish I had the
1:11:37liquidation data but I
1:11:48do let's just see what it does first I'm
1:11:51just curious I'm going to put that on
1:11:53the Note sheet though put Liquidations
1:11:56in
1:11:57there these could be good to help
1:12:02identify points of
1:12:06capitulation
1:12:07upward or down okay because I got
1:12:12volume um I could also put things like
1:12:15open
1:12:18interest inter
1:12:21interest funding rates okay now we going
1:12:25now we got the wheels turning
1:12:35B open interest what
1:12:38else okay that's good for now
1:12:43um here's my
1:12:47code 144 lines look at us look at us
1:12:51go there it is
1:12:55all right let's uh let's let it
1:12:59run I'm gonna make
1:13:05a this is the end though I'll see you
1:13:07some other time this is long enough we
1:13:10I've shown you enough we are diving into
1:13:12unsupervised machine learning for
1:13:14trading does machine learning work for
1:13:17trading can you predict
1:13:19price well I'm sorry to not bury the
1:13:22lead but no you can't predict price
1:13:25maybe maybe if you got some hidden model
1:13:27but the way I look at it from testing
1:13:30this stuff is if we're all using similar
1:13:33models they're all available with Claud
1:13:36or gbt or whatever if we're all using
1:13:39the same models to predict price well
1:13:42that price will be predicted then by you
1:13:46you that dude over there your grandma me
1:13:51and then that price will not be the
1:13:52price anymore because it's already been
1:13:54predicted by thousand thousands of
1:13:56people so thousands of people run these
1:13:58models I don't know how many models
1:14:00there are 20 30 50 100 whatever like we
1:14:04tested them all right
1:14:07collectively predicting price the next
1:14:09price I don't know if that's the way
1:14:11maybe though maybe if you build some
1:14:13unique model I don't know I'm kind of
1:14:16thinking thinking thinking this through
1:14:17with you as always because that's what I
1:14:19do here I just show you everything I'm
1:14:21doing
1:14:22live and
1:14:25what this has led me to is well what was
1:14:27Jim Simons using well Jim Simons he was
1:14:29using hidden Markoff models okay so
1:14:31let's build a hidden Markoff model and
1:14:33then let's watch everything about Jim
1:14:35Simons okay that's what we did that's
1:14:36what we've done that's what we're doing
1:14:38I've got a bunch of notes here and a
1:14:40bunch of ideas and we
1:14:43have a couple models here and you can
1:14:46see we even tested some out of sample
1:14:50data so we have multiple models here I
1:14:53think this was the template model so so
1:14:55this was the first hmm with just putting
1:14:59in the returns volatility and volume
1:15:00change okay and then the next one we put
1:15:03in as the features we put in these
1:15:06features here so we added some more
1:15:09returns volatility volume change BB
1:15:11width RSI and ema2 okay and then this
1:15:14one we tried some other things you can
1:15:17see the states we're trying to predict
1:15:19we're trying to predict seven different
1:15:20states but the thing with hidden Markoff
1:15:22models so far and excuse me if I or any
1:15:25of this I'm not I'm not like a gilfoil
1:15:27type senior engineer here I'm a womb
1:15:29engineer I'm still in the womb
1:15:32so bear with me gil foil but the thing
1:15:35about hmm is it doesn't actually know
1:15:38what the state names are so we actually
1:15:39have to rename them after it finds the
1:15:41states it finds different states
1:15:43different regimes and that's kind of
1:15:45what Jim Simons was doing supposedly but
1:15:47do you think he'd actually tell us what
1:15:48he's doing no way but maybe we could put
1:15:51the pieces together that's my my
1:15:52hypothesis here we we are just passing
1:15:54in returns volatility and volume change
1:15:57top three Indies these were the top
1:15:59three ones volume change is taking up
1:16:01most of it though and how can we know
1:16:02that well when we run this we can see
1:16:05that it's mostly getting weighted to the
1:16:07volume change I'll show you some cool
1:16:08things here first let's show the plot
1:16:11it's making so it's fitting the model
1:16:13this might take a
1:16:18second take a little bit of time it's
1:16:20doing its little AI thing machine
1:16:22learning
1:16:23thing it's it's fitting the
1:16:26model boom all right saving plot perfect
1:16:29perfect and it also plotted the plot so
1:16:31you can see these are the different
1:16:33regimes here it colored them but the
1:16:35thing is it doesn't know the names of
1:16:37these regimes so in reality we got to go
1:16:39back and we got to set up these regimes
1:16:46ourselves I'm sorry
1:16:49baby I've got a dog here that constantly
1:16:52stands right under my foot so I almost
1:16:54just killed K her that was very very
1:16:56scary you okay baby
1:16:58girl I didn't put my full weight on her
1:17:01thank God that I know she does that
1:17:04because she likes to I I stand on one
1:17:06foot a lot of the times while I'm coding
1:17:08and then she'll go right under my my
1:17:10other foot that I come down on and I'm a
1:17:12big boy you know that's a big that's a
1:17:15lot of way for a little dog
1:17:17so I almost had a heart attack right
1:17:20there I'm sorry for the for the side
1:17:22show but this is life and I'm sure you
1:17:25got kids or dogs and things like that
1:17:28but anyways here you can see the seven
1:17:31different regimes we have bullish
1:17:32bearish sideways but like I said it
1:17:34doesn't know the regimes it just cuts up
1:17:36the data this is Bitcoin data here for a
1:17:39few years and now you can see this
1:17:42outputs
1:17:44the the analysis the analysis has been
1:17:48completed not fully though you can see
1:17:50the state prediction is 89%
1:17:55but we already know all this I'm just
1:17:56doing a little review because again this
1:17:58is fairly new to me as well and every
1:18:00day I just go live and show you what I'm
1:18:02working
1:18:07on so it's still doing some other things
1:18:10here and I can show you the final
1:18:12results we have thus far if I go to the
1:18:15read me I've been tracking them
1:18:19here so down here
1:18:23we you can see this hmm the seven
1:18:28states it's our first attempt and it had
1:18:3287% State prediction accuracy the log
1:18:35likelihood was negative 15,000 you can
1:18:37see this one is done now and this one is
1:18:41uh you can see what I was trying to show
1:18:42you is volume change is taking up all of
1:18:44the importance can you explain this so
1:18:47I'm I'm using H&M mm hmm with three
1:18:51different
1:18:52features and it converges to volume
1:18:55change being the most important
1:18:57feature I'm using an hmm and this is the
1:19:03output can you explain can you
1:19:07explain what the
1:19:10importance means in my mind this means
1:19:13that this is the most important feature
1:19:14it's pretty self-explanatory but I just
1:19:16want to make sure that I'm correct the
1:19:19feature importance analysis you provided
1:19:21shows the relative importance of
1:19:23different features in your hidden Market
1:19:24off model let me break down what these
1:19:27numbers
1:19:28mean volume change 94.5
1:19:321% Bullinger band width
1:19:372.72% volatility
1:19:412.77% these numbers represent the
1:19:44estimated contribution of each feature
1:19:47to the model's prediction or performance
1:19:50the value sum up to 100% indicating the
1:19:53relative importance of each feature in
1:19:56your case volume change is by far the
1:19:58most important feature accounting for
1:20:0194.5% you can see that here it's 0945 so
1:20:04we just got to do a little
1:20:06math not really but of the model's
1:20:09predictive power this suggests that
1:20:11changes in trading volume are the
1:20:14primary driver of the patterns or States
1:20:17your hmm is
1:20:19detecting volatility and BB width which
1:20:22I assume stands for Binger band wi have
1:20:26much lower importance this analysis
1:20:28suggests that hmm is relying heavily on
1:20:30the volume changes to make its
1:20:32predictions or determine State
1:20:35transitions while other two features
1:20:37play a much smaller role it is important
1:20:39to note that feature importance doesn't
1:20:41necessarily indicate
1:20:44casuality cause causality sorry but
1:20:48rather how much the model relies on each
1:20:51feature for its predictions High
1:20:53importance could mean mean the feature
1:20:55is truly more relevant to the problem or
1:20:58it could indicate that the feature is
1:21:00easier for the model to use in
1:21:15predictions so what I want to do here is
1:21:17I want to look at some of our
1:21:19notes and these notes are going to show
1:21:22us the differences and one of the tests
1:21:25I did was
1:21:27since you can see here uh where is it I
1:21:30think it's one of my last
1:21:32ones I said no volume since volume
1:21:36continuously took up 90 some per 94%
1:21:40here of the importance 93% of the
1:21:44importance here and 97% here I took it
1:21:48out I took it out because I wanted to
1:21:50see something a little more a little
1:21:52more
1:21:55what do you call it balanced yeah a
1:21:57little more balanced so I took it out
1:21:59here and you can see this made it way
1:22:01more
1:22:04balanced and you can see it actually
1:22:06increased to 95% here so it was before
1:22:1189% the state prediction accuracy and
1:22:14that prediction accuracy is how accurate
1:22:17is it changing states in predicting
1:22:20changing states so when the CH colors
1:22:22change how accurate is it in and uh
1:22:26deciding predicting the
1:22:31change and again I'm not the best in the
1:22:33world at this stuff guys so you know
1:22:35bear with me I'm kind of showing you as
1:22:37I learn as always that's what this whole
1:22:39channel is so every single day I just go
1:22:41live and I try the hardest things and
1:22:44try to understand it uh I was so scared
1:22:46to learn how to code when I got started
1:22:47dude I I waited 15 years I was in t
1:22:50typing on this computer fully in Tech
1:22:53without knowing how to code for so long
1:22:55dude just cuz it's scary but I now I'm
1:22:57at the point of life where I need scary
1:22:59things I need hard things so I just go
1:23:01after them viciously every single day
1:23:02here live so then the the common thing
1:23:07was or the obvious thing to do was Let's
1:23:09test all of these models I think there's
1:23:11five different models on out of sample
1:23:15data data that the model has never seen
1:23:17before and you can see here are the
1:23:19results
1:23:30state zero so these are the two best
1:23:34ones I have so far one has volume change
1:23:39one
1:23:40doesn't and these are the
1:23:43results says model one has log
1:23:46likelihood of
1:23:47negative 65 6956 and this one has
1:23:533890 so I've got these two models
1:23:57explain these
1:23:58datas I've got uh I tested I tested two
1:24:02different uh hmm
1:24:06models and the code uh on out of sample
1:24:10data models on
1:24:12oos and the code is attached below are
1:24:18the results and the
1:24:22plot please watch walk me through
1:24:25them so I understand okay let's put the
1:24:29code the code in there as well copy okay
1:24:34and here's the
1:24:38code okay and then also I want to put
1:24:40this screenshot in there and do you
1:24:42think I saved it
1:24:43here
1:24:45no I did
1:24:48not I just want it to to have
1:24:51everything everything I got and I know I
1:24:54already reviewed this actually yesterday
1:24:56but dude this is the only way I can
1:24:58learn is with
1:24:59repetition 500 shots a day baby 500
1:25:02shots a day I've been doing that since I
1:25:04was a kid come on you think I'm not just
1:25:06going to do the same thing over and over
1:25:08again till I learn it of course I am
1:25:10worked for me in the past it's going to
1:25:12work for me in the future repetition is
1:25:14key just keep attacking every single day
1:25:17four hours a day 4 hours a day if you're
1:25:20not doing 4 hours a day then don't talk
1:25:21to me don't ask me me questions because
1:25:26all your questions can be answered in
1:25:28four hours a day we got AI my guy don't
1:25:30tag me don't DM me nothing if you're not
1:25:33putting four hours a day of focused work
1:25:36at this goal to automate your trading
1:25:39don't hit me about nothing about
1:25:41automated trading nothing I'm sorry dog
1:25:45we're competitors at the end of the day
1:25:47but I'm showing you every single thing
1:25:48cuz I believe code is a great equalizer
1:25:50for me it is at
1:25:52least can't stop me now dog I know how
1:25:55to code I can build anything you can
1:25:56build anything we got AI my
1:25:59guy thank you for providing the code and
1:26:01results I'll walk you through the key
1:26:03aspects of your hidden Markoff model
1:26:06analysis let's get these up model
1:26:09comparison you've tested two hmm on out
1:26:12of sample data model one uses Binger
1:26:15bands volatility and RSI as features
1:26:18model 2 uses volume change Binger band
1:26:21width and volatility as features
1:26:26so the log likelihood
1:26:27scores did I write this down in the out
1:26:31of sample data no I didn't
1:26:34unfortunately that's okay we'll write it
1:26:36down right
1:26:40now so model 1 has - 6956 model 2 has
1:26:453890 the log likelihood is a measure of
1:26:48how well the model fits the data higher
1:26:50values indicate better fit
1:26:58model 2 has a significantly higher log
1:27:01likelihood suggesting it fits out of
1:27:03sample data much better than model one
1:27:06visualization the plot shows three
1:27:09panels top panel bitcoin price over time
1:27:12this gives context on the market
1:27:14conditions during the out of sample
1:27:22period middle panel shows the I need to
1:27:25get this up if we're going to I need to
1:27:27get the
1:27:29um what they're talking about up or else
1:27:32we can't understand it can't understand
1:27:35anything you can't see even if it's in
1:27:36your mind B you can see it in your mind
1:27:39your mind doesn't know what's real or
1:27:41not because your mind is
1:27:44real so if you can see something in your
1:27:46mind you can have it I believe I can
1:27:50fly I believe I Could Touch the Sky
1:27:55I think about it every night and
1:27:59day spread my wings and fly
1:28:03away I believe I
1:28:06[Music]
1:28:07can I see me running through that open
1:28:13door I Believe I Can
1:28:17Fly
1:28:19D you got to believe it before you do it
1:28:21D the log likelihood okay here we go um
1:28:27top panel is the Bitcoin price over time
1:28:29just so we can see it this is the out of
1:28:31sample data here middle panel model one
1:28:35States shows how model one classifies
1:28:38different time periods into seven
1:28:40distinct States 01 through
1:28:44six each color represents a different
1:28:46state
1:28:54bottom panel model two State
1:28:57similarly shows State classification for
1:28:59model two this is interesting so while
1:29:02the price is here you can see how the
1:29:04state changes it's kind of hard to see
1:29:07to be real
1:29:08but State transition both models show
1:29:11frequent State transitions indicated by
1:29:14the changing colors in the lower two
1:29:17panels model 2 appears to have more
1:29:20distinct and longer lasting State
1:29:22periods which might suggest it's
1:29:24capturing more persistent Market
1:29:27regimes feature importance given that
1:29:30model 2 performs significantly better it
1:29:33suggests that volume change which is in
1:29:37model 2 but not in model
1:29:43one is a crucial feature for predicting
1:29:46bitcoin price
1:29:48States this aligns with the importance
1:29:51feature analysis you shared earlier
1:29:54where volume change had the highest
1:29:55importance model stability model 2 seems
1:29:58to produce more stable State assignments
1:30:02longer periods of consistent
1:30:05colors implications implications the
1:30:08superior performance of model 2 suggests
1:30:11what volume changes are more
1:30:14import informative for predicting
1:30:16bitcoin price than
1:30:21RSI the frequent State changes in both
1:30:25models reflect the volatile nature of
1:30:27cryptocurrency
1:30:30Market to further understand the models
1:30:33you might want to analyze the
1:30:34characteristics of each
1:30:38state interpret look at transition
1:30:40probabilities between states consider
1:30:43how you might use these State
1:30:45predictions in trading so
1:31:02analyze the characteristics of each
1:31:03state okay how can I make this how can I
1:31:06make this more
1:31:09robust and test
1:31:13more
1:31:15features I was
1:31:18thinking I can add in more indicators I
1:31:23have un
1:31:24limited from pandas
1:31:27ta um and ta
1:31:31Li I was also
1:31:35thinking about trying just four
1:31:41states and what were those four states
1:31:43somebody shouted it out the other day so
1:31:45much love to
1:31:46them and those four states he said that
1:31:49Jim Simons looked at
1:31:57we those four states dude let's go look
1:31:59at our notes Here notes are quite
1:32:16dope four volatility regimes
1:32:32this was the
1:32:34comment on four
1:32:40states okay let's just let's kind of IDE
1:32:44ideate
1:32:48ideate yo 777 Smokey what up bro what
1:32:52does each state mean the different
1:32:53regimes
1:32:54uh you were flying bro I saw it
1:32:57happen
1:32:59uh puppy is
1:33:02Gucci Gucci Gucci bless you thank you
1:33:04thank you thank you apple apple apple
1:33:08little apple little
1:33:14mananita okay your ideas for improving
1:33:16the robustness of your hmm model are
1:33:19excellent let's explore how excellent
1:33:21you gave me excellent hey shorty you
1:33:23ain't never said hey I'm excellent dog I
1:33:25don't I don't care I'm not I'm new to
1:33:28this game dog and I already got
1:33:29Excellence from a a AI my guy is this
1:33:33excellent is this am I excellent dog I
1:33:36did not know that that is the biggest
1:33:38compliment I could ever get from CLA I
1:33:40think I'm on CLA forever yo they
1:33:42probably coated that in they're like yo
1:33:44juice them up juice them up hey hey hey
1:33:47if moev ever gets on claw juice him up
1:33:49say he's excellent all the
1:33:51time that's crazy excellent I never
1:33:54heard any AI tell me excellent okay
1:33:57let's explore how you can enhance your
1:34:00approach excellent GE your boy ain't
1:34:03excellent I'll tell you that much we did
1:34:05the math the other day I've been at this
1:34:07for three and a half years 3.5 * 365
1:34:11days time. 85 because I took some time
1:34:14off 1,000 hours Time 4 hours I'm only at
1:34:184,343 hours I ain't excellent bro you
1:34:21know who excellent is Jim Simon's
1:34:23excellent rest in peace my
1:34:31guy let's explore how you can enhance
1:34:34your
1:34:35approach testing more features
1:34:37leveraging panas ta and ta lib gives you
1:34:39access to a wide range of technical
1:34:41indicators here's how you could approach
1:34:43this add more indicators Trend
1:34:44indicators moving averages SMA EMA magd
1:34:47P parabolic star momentum indicators
1:34:50stochastic oscillator CCI volume
1:34:53indicators on
1:34:55balance cool cool ATR nice keler
1:34:58channels nice feature selection after
1:35:00adding more indicators you might want to
1:35:02use feature selection techniques to
1:35:04identify the most informative ones
1:35:06consider using methods like correlation
1:35:10analysis
1:35:11PCA random Force
1:35:21importance try and force state Jim
1:35:24Simon's insight about four volatility
1:35:26regimes is valuable here's how you can
1:35:28implement this modify your hidden hmm
1:35:31initialization to have four components
1:35:33instead interpret the four states you
1:35:36could potentially interpret these four
1:35:37states as low volatility Rising
1:35:39volatility High volatility yo that's F
1:35:41four that's four making the model more
1:35:43robust cross validation instead of
1:35:46single train test split use time
1:35:50series cross validation to ensure your
1:35:52model performs well
1:35:54across different time periods Ensemble
1:35:57methods consider creating an ensemble of
1:36:00hmm with different feature sets or
1:36:06parameters regularization if overfitting
1:36:09is a concern you can add regularization
1:36:12while hmm doesn't directly support this
1:36:16you can implement it by adding a small
1:36:18constant sensitivity analysis test how
1:36:21sensitive your model is to changes in
1:36:23the input features or parameters this
1:36:25can help you understand which aspects of
1:36:27your model are most crucial back testing
1:36:29Implement a back testing framework to
1:36:31evaluate how well your hmm States
1:36:34predicted predict future price movements
1:36:36or
1:36:38volatility okay so these are pretty good
1:36:40ideas
1:36:46honestly I'm only saying that because
1:36:48she hyped me up though you know it oh
1:36:51yeah you got some pretty good ideas is
1:36:53gbt or
1:36:59Claudia but in real you know is just cuz
1:37:02she hyped me up and you know I'm a
1:37:04lifetime I'm a lifetime now I'm a
1:37:07lifetime now to do next next things I
1:37:12can do to improve
1:37:16hmm dope dope
1:37:19dope okay so get him out of here get
1:37:22that BTC little boy out of here 150
1:37:26racks 150 racks I hope that's not you
1:37:29dude I hope you're here instead tap in
1:37:33777 Claud B
1:37:35glazing yo that's fire
1:37:38B I don't even I've never heard that in
1:37:40my life but she do be glazing be CLA be
1:37:46glazing y'all keep me hip I'll tell you
1:37:49that much y'all keep me hip CLA be glaz
1:37:53in be I'm about to start saying that all
1:37:56the
1:37:57time she be glazing be I be glazing
1:38:02too let's just double check make sure
1:38:04that's not something bad Urban de
1:38:08dictionary glazing the act of f
1:38:13no when you are meat riding someone
1:38:19oh I don't be doing that I don't be
1:38:22doing that I be doing tomor so it's me
1:38:25kby glazy okay okay okay see this is old
1:38:29man old man trying to be hip over here
1:38:32he trying to be hip saying the word hip
1:38:34ain't hip dog she be glazing that's good
1:38:37I need to be
1:38:38glazing all the time that's how she
1:38:41going to keep my subscription she better
1:38:43keep glazing this
1:38:46she hey what year were you born dog
1:38:49that's crazy that's crazy I love it I
1:38:52love I love how Lang language
1:38:54changes I love it I love how slang
1:38:57changes and I feel like we're going
1:38:59through a pretty big shift right
1:39:02here for us old heads yo Brian says
1:39:06hello sir I wanted to ask the mentorship
1:39:10I can use code build trading Bots like
1:39:13Jim Simons you got to you got to build
1:39:14it yourself for sure but um the boot
1:39:17camp I think is what you're asking about
1:39:20the boot camp will help you I've been
1:39:22doing this for how many hours did we
1:39:23just say I've been doing this for 4,300
1:39:26hours so I have a boot camp that I try
1:39:28to just share the things that I know
1:39:32most and the things that work for me and
1:39:35the pieces to the puzzle but at the end
1:39:37of the day everybody's puzzle is going
1:39:39to be different Jim Simon's puzzle is
1:39:41going to be much different than my
1:39:43puzzle my puzzle is going to be much
1:39:44different than Brian's puzzle Jane's
1:39:47puzzle is going to be different than
1:39:49Brian's puzzle so it all comes down to
1:39:52your unique Ed your strategy your
1:39:54approach to the markets but I can show
1:39:56you everything in between step by step
1:39:59how to automate your trading how to test
1:40:01if it worked in the
1:40:02past gives you a better hope that it's
1:40:05going to work in the future but it's not
1:40:08guaranteed and uh yeah everything's in
1:40:10the boot camp 69 bucks stupid cheap high
1:40:15vow you can read what others have to say
1:40:18that's
1:40:21it landlord says are we continuous in
1:40:23the hmm from yesterday
1:40:26yeah yeah dog yeah
1:40:39we send me back send me a simple
1:40:50script that will print out all of the
1:40:56indicators for
1:40:59pandas ta and ta
1:41:05live yo Brian says saw you your channel
1:41:10and you're a King sir I appreciate you
1:41:11dog that is so much love much love to
1:41:13you dude glad you're here I'm so so
1:41:16happy you're here let's say uh all
1:41:20indicators
1:41:21all sh
1:41:23no all
1:41:26indicators
1:41:28please AI my guy ai
1:41:33ai dude this is over it's over for me
1:41:36and you all of the indicators yo all of
1:41:40the indicators you see that dude you
1:41:42want that script come with me dog
1:41:45I'll show it every day dog I'll show you
1:41:49everything
1:41:51everything output them to a
1:41:54output to a txt in my data folder yo I
1:41:59don't care who you are dog I don't care
1:42:01if you've been coding for 30 years dog
1:42:03I'm coming at your neck every single day
1:42:06I got the power now we got the AI my gu
1:42:10stop
1:42:11it stop it do you see this it's just one
1:42:15idea and boom I get every single
1:42:18indicator just like that I don't need
1:42:20some long tutorial I don't need a c
1:42:22stack overflow I don't need to go to
1:42:24school for 82 years like you did dude
1:42:27we're not the same I go harder than you
1:42:29you can't keep up I'm
1:42:32sorry yo print them out too
1:42:35cousin whatever that's
1:42:39cool so these are all the indicators
1:42:46here for ta lib pandas ta
1:42:50indicators so we have
1:42:54all these I'm going to use adx yeah
1:42:57let's do that let's do that so I'm going
1:42:58open this to the side
1:43:00here split right show all indicators
1:43:27sorry burped in your ear my bad that was
1:43:29super in
1:43:37poite yo Mev you need to start using
1:43:40multi-processing and multi-threading
1:43:43why it's good comment
1:43:47though it's good comment but why
1:43:51do especially when accessing large sets
1:43:54of
1:43:56data okay okay I seen somebody play with
1:43:59that the other day up in the
1:44:02Discord just so I can test faster back
1:44:05test faster multiple times
1:44:08multiple multiple processors I
1:44:12guess I like the idea I appreciate the
1:44:15share much love to you for that you can
1:44:17use Python and
1:44:22C++
1:44:52e for
1:45:28all right so I'm going to go ahead and
1:45:30start writing some of these ideas
1:45:33now so test two test
1:45:41three
1:45:43Indies to
1:45:45try so I want to try the adx because
1:45:49homegirl talked about it a lot um one
1:45:52from my first podcast I
1:45:55did uh Mansion says I use multi
1:45:59processing it's much faster to train the
1:46:02model okay there we go that's a good why
1:46:05that's a good why that's on
1:46:07track use
1:46:11multi-processing to train models and
1:46:15also to back test faster appreciate you
1:46:18guys this is like you see that right
1:46:20there like it might have taken me two
1:46:22years to get there and y'all just
1:46:24dropped a dime like that you just Dro
1:46:26you just dropped Mev a dime you threw
1:46:28the alleyoop and I just dunked it I
1:46:31haven't dunked it yet but I'll dunk it
1:46:33soon landlord says he wishes he had a
1:46:35supercomputer Connor says uh was using
1:46:39my uni one for the last couple years and
1:46:41I miss it dearly dang dude what's the
1:46:43super computer looking
1:46:48like Connor says plus one for the faster
1:46:51training thank you landlord
1:46:53thank
1:46:55you appreciate all of your your
1:46:57Brilliant Minds I don't know a lot I
1:47:00don't know most honestly I only have
1:47:034334 hours in this game so far and
1:47:08um quantum computer dang I got the I got
1:47:12the quantum MacBook no I'm just kidding
1:47:15let's see what other indicators we want
1:47:17to try I'm just going to use ones that
1:47:20are um
1:47:26I
1:47:28guess I don't know that's the hard part
1:47:31like which ones do you use
1:47:34adx try that I just the ones that I
1:47:38guess I know
1:47:42ATR there's just so
1:47:51many Don
1:47:54Chan I'm just based off memory of tests
1:47:57I've done in the past or ideas I've
1:48:00heard I'm just going to use
1:48:04those I
1:48:08guess and if you got any ideas that you
1:48:11see here holler at your boy just be like
1:48:14yo Moon try that
1:48:18one Lyn regression
1:48:36macd yo how much like RAM and stuff does
1:48:39a supercomputer have I'm just curious to
1:48:41see like yeah how far off I am I know I
1:48:44don't have anything near a supercomputer
1:48:46but like what is yeah cores 2 I guess
1:48:50again I'm not very I'm not very
1:48:51technical I'm just kind of a new
1:48:53but appreciate y'all yeah I want to Cur
1:48:56I'm curious how far off I
1:48:59am prom super computer status what is
1:49:03out there and then can we use gcp Google
1:49:06cloud or GC
1:49:08whatever to get supercomputers is it
1:49:11worth it I'm curious about all that
1:49:14stuff but this is fun this is fun true
1:49:21range I love this game game dog this is
1:49:23just this is my element here feature
1:49:25engineering this is where I'm at this is
1:49:28this is my game right here come up with
1:49:29ideas I love it I love it I love it ta
1:49:43live they got all the math
1:49:49here cosign seal
1:50:00RSI Stoke Stoke RSI how about
1:50:11that what's
1:50:13Kama
1:50:15Kamala Kamala
1:50:25CDL CDL engulfing what are these I don't
1:50:28even know what these are D CDL is chart
1:50:31something doy
1:50:35Hammer let's put it let's just try it I
1:50:37don't know dude I'm just trying
1:50:39everything dude this is the game right I
1:50:42don't know we'll see see I mean we'll
1:50:45see if it's better or
1:50:47worse
1:50:49ly standard DV T range ATR we got the
1:50:54ATR in there already the
1:50:57VAR the
1:50:58obv you down with obv yeah you know me
1:51:03candlesticks CDLs got it thank you bro
1:51:08um so Connor says it was mainly the GPU
1:51:11access training models was like day and
1:51:13night when looking at using
1:51:16supercomputer versus
1:51:17not angga says you can use tpus tensor
1:51:21processing units in Google collab nice
1:51:25landlord says you can run your scripts
1:51:26on AWS and
1:51:28gcp but you pay for the increase in Ram
1:51:32I think I'll stick with gcp for now if a
1:51:35Candlestick is
1:51:36100 it's bullish if it's negative 100 is
1:51:39bearish good Intel
1:51:51bro for
1:52:25you good I didn't kill you over there
1:52:27did I okay let's try this um I'm going
1:52:29to
1:52:30say top three Indies no no no do not
1:52:35there top three
1:52:37Indies is this use the volume change
1:52:40okay let's say update this
1:52:43code update the attached code to not use
1:52:49the features or the
1:52:52yeah
1:52:54features it
1:52:56currently has but use the ones
1:53:01below from pandas ta I think that's the
1:53:05first one
1:53:07right let's try these
1:53:10three yo is this is this a approach is
1:53:13this how people approach this stuff or
1:53:15are there like smarter people out there
1:53:24I found that three is better than seven
1:53:31so Myra says you come from data science
1:53:36background Munda I
1:53:40mean I don't know not really not at all
1:53:44actually so I learned how to code 3 and
1:53:47a half years ago but I learned through
1:53:49watching like 10 machine learning
1:53:52courses
1:53:53so I've seen everything not everything
1:53:55but I've seen a lot of machine learning
1:53:58stuff I was just super fascinated by it
1:54:00but I also was learning how to code at
1:54:02that same time so I was just typing all
1:54:03this machine learning code out in Python
1:54:06and kind of learning python at the same
1:54:08time I don't think it was the correct
1:54:09way to do it but I've seen it all now
1:54:11and now I'm three and a half years
1:54:13deeper and I actually know how to code
1:54:14in Python I can build any Bots anything
1:54:16like that build a bunch of alos that I
1:54:18run every single day and now I'm diving
1:54:21back into machine learning so I think
1:54:22it's going to click it has been clicking
1:54:24a little easier so if you want to say
1:54:27that's a background then yes but no I
1:54:29didn't go to school for any of this
1:54:30stuff
1:54:32dog I did not go to school for coding
1:54:34machine learning
1:54:36nothing I didn't read a lot of these
1:54:39boring books and try to absorb what I
1:54:44can so all this all everybody commenting
1:54:47here you guys are Lifesavers bro
1:54:50teaching me what you know
1:54:52because I know we got some guil foils in
1:54:54the
1:54:55audience Connor says for individual
1:54:57sessions I could run I could request up
1:55:00to 72 CES
1:55:05dang wow so 72 time eight I can't do
1:55:09that math
1:55:13bro all right so I'm way off I got 96
1:55:16gigabyt of RAM on my computer though
1:55:23altered says I want to enter your
1:55:26community I don't speak English is it
1:55:29possible to add French language or
1:55:31Italian dude you literally just said
1:55:34that in English you do speak English
1:55:36give yourself some credit first
1:55:39off but yeah
1:55:43um yeah everything's got subtitles for
1:55:45you bro so you can pick French or
1:55:47Italian that's cool you know French and
1:55:50Italian that's amazing but yeah you can
1:55:52join the boot camp everything in the
1:55:54boot camp has subtitles to every
1:55:56language pretty
1:56:01much so that's dope uh Myer says that's
1:56:05cold straight to ml had to go for the
1:56:08top man had to you know that's that's
1:56:10who I got inspired by it seems super
1:56:14hard super fascinating too and it was
1:56:17before AI was
1:56:19here well before llms were usable from
1:56:24the desktop like they
1:56:27are so I'm glad I learned it kind of by
1:56:33him before I just demand AI to do
1:56:39everything all right I just had to
1:56:42finish my little Waffle
1:56:47there what do we want to call this one
1:56:50adx ATR
1:56:52Don
1:56:55chin adx
1:56:58ATR
1:57:02Don I like that adx ATR Don
1:57:06Chim let's just pop it in let's pop the
1:57:09code
1:57:11in dump it too yeah dude dump it I want
1:57:18that update this though I want to say
1:57:21train models
1:57:23here
1:57:25data I want to make a folder this make
1:57:27this a little
1:57:28neater
1:57:30trained
1:57:32models and let's put all of these models
1:57:38here copy
1:57:43path update the code so
1:57:48it I'm not going to do that
1:57:54just going to do it myself
1:57:56dude did I put it in there already dang
1:57:59that boy
1:58:01nice that boy
1:58:04nice look at this I can type code in the
1:58:07in the thing
1:58:09dude all right let's see what it
1:58:13does run it do fit in
1:58:21fitting somebody just got licked we got
1:58:24licks always let's see what's going on
1:58:26today oh my
1:58:28God people getting wrecked
1:58:32today I'm looking at the 15minute chart
1:58:34that's why this taking a long time to
1:58:36fit or
1:58:38whatever there it is there we
1:58:43go
1:58:48so be in the market in green and out of
1:58:51the market outside of the green I like
1:58:54this in the market green out of the
1:58:57market outside
1:59:00green yo call trading what up bro oh is
1:59:04JPY going crazy again what's going on
1:59:06over
1:59:11there Connor was training models to
1:59:14predict gra gravitational waveforms of
1:59:17binary black hole collisions on the
1:59:19supercomputer very resource intensive
1:59:21stuff
1:59:22not as applicable as I'll go trading
1:59:25though oh man that sounds intense
1:59:29though okay so this is interesting data
1:59:32so let's go ahead and analyze
1:59:39it we just got to do repetition dog I
1:59:42don't know where you're at in your
1:59:43journey but for me I just need reps I
1:59:46need reps reps reps reps reps reps
1:59:5798% likelihood
2:00:05wow I'm curious about the importance
2:00:08here
2:00:24count okay that's fine I was like why
2:00:26are they all the same it's still working
2:00:30here nice to see you here again Call of
2:00:34Duty what is
2:00:37happening is the expectation over the
2:00:40September meeting in fomc 25 bips versus
2:00:4350 bips and inflation is running
2:00:45relatively high in Japan which leads
2:00:47them to have to increase rates they will
2:00:50do so slightly but what matters most is
2:00:52future
2:00:54expectations over falling rates in the
2:00:56US
2:00:57interesting okay look at this I like
2:00:59this too because this uh see how it's
2:01:03pretty even I don't know if that matters
2:01:12but and compare it to our best model
2:01:18below and what was our best model below
2:01:20is the question damn
2:01:29um hey it didn't update into our
2:01:34models
2:01:40uhoh ibly just overwrote it
2:01:48huh h
2:01:53I put it in the data folder yeah my
2:02:08bad it's typically me that's what I love
2:02:11about coding when something goes
2:02:14wrong it's typically
2:02:17me it's like such great accountability
2:02:25and then I want to compare it
2:02:33to let's see where are our other hmm
2:02:36models here noall dang I don't know how
2:02:40to tell which is the best one I can look
2:02:42at the out of sample so out of sample
2:02:45data
2:02:47here and X2 X1 so X 2 model 2 is the
2:02:52best volume change BB width and
2:02:57volatility Vol BB and
2:03:08VA I love this game dog you'll never get
2:03:11me to stop I'm sorry
2:03:23you'll never get me to stop this game
2:03:25I'm sorry BB volatility and
2:03:29RSI I'm sorry dude I'm not going
2:03:32anywhere for 60 years at least I'll be
2:03:34standing my black ass right here for 60
2:03:37years at
2:03:38least BB width
2:03:42volatility and RSI
2:03:55how do you evaluate Indie importance I
2:03:58don't know dude it says right
2:04:02here it's a good question
2:04:07though um and then compare it to our
2:04:10best model which is model two so I just
2:04:14don't I don't know any of this yet I'm
2:04:16I'm getting there though I'm getting
2:04:17there I'm starting to understand a
2:04:18little better a little bit better every
2:04:21day every day
2:04:33everywhere uh it's Jan says is each
2:04:36console log a unique trading bot
2:04:40no it's
2:04:42not right now I'm working on the hmm
2:04:45which is a hidden markof model which
2:04:46supposedly Jim Simons liked but you know
2:04:49that boy capping
2:04:53ain't nobody going to tell us what that
2:04:54was actually
2:04:55working but we did figure out yesterday
2:04:58that he definitely starts with data
2:05:01first and if you start with data first
2:05:02that means that you're not looking at
2:05:03indicators and stuff like that you're
2:05:05looking at the data so we're going to
2:05:06let the data be do it Thug fizzle you
2:05:09know we're going to start with data
2:05:11First Data first and that's what we're
2:05:13doing now we're going to we're sending
2:05:14into the the AI the machine learning
2:05:17what's this JPY you're talking about I
2:05:19don't even know how to get there let's
2:05:20see if I can get there JP
2:05:24I I'm not a I'm not a Forex guy I
2:05:28see oh boy is this what's this
2:05:33mean what does up mean is that bad or
2:05:37good because I know with these pairs and
2:05:41stuff cuz the other day didn't it like
2:05:44okay the JP I capitulated down
2:05:49here honestly I do not know
2:05:55I need to get a little better at
2:05:56currency understanding I
2:05:59think cuz it looks like it's
2:06:02important but see if this is done
2:06:06yet being I feel like you need to
2:06:09understand currencies and how they
2:06:17move USD JPY
2:06:21oh so this was
2:06:29opposite oh that's a different
2:06:32story so was that just
2:06:38backwards did I just look at it
2:06:42backwards this is what yeah okay yeah
2:06:45this crashed the other
2:06:46day but that doesn't look like a
2:06:48backwards chart does it
2:06:50JPY I don't
2:06:52know JPY
2:06:55USD uh that's not backwards though is
2:06:58that a different
2:07:00thing sorry for my ignorance again
2:07:02appreciate you guys' love
2:07:06though okay
2:07:08so it's the
2:07:11inverse okay so this down bad today huh
2:07:15I should add this to the list
2:07:18here what are some other good ones I
2:07:20should start looking at that like move
2:07:23the
2:07:24economy let me just add them to the list
2:07:26because you know I canceled this you
2:07:28know I canel the uh trading view there's
2:07:31no reason for it what are some good
2:07:32symbols I should watch in order to get
2:07:35more in tune with Forex because I know a
2:07:37lot of you want
2:07:38Forex over here and um I've heard that
2:07:42Forex is more mean reverting so it could
2:07:45be interesting to look
2:07:46into but let's go ahead and analyze this
2:07:50data so we got two new models done or
2:07:53one new
2:07:54model and I want to keep just working
2:07:57through this until it just so so
2:07:59clear okay yo thank you for this so we
2:08:03have we've got some good suggestions as
2:08:06well I know I'm always doing this
2:08:10but always getting over here so Euro
2:08:15USD that makes
2:08:20sense just got to get them on the uh the
2:08:24thing before it cancels on
2:08:26me so Euro
2:08:28USD what else GP GBP GBP the British
2:08:34pound GBP
2:08:39USD and
2:08:41USD
2:08:43JPY those are the main three sweet yo I
2:08:46appreciate you
2:08:48dude caller trading and landlord I appre
2:08:51appreciate both of
2:08:53y'all so these are the three
2:08:59movers so Euro
2:09:10USD
2:09:12GBP so the British
2:09:15pound who uses the British pound Britain
2:09:28oh the UK okay so Jane uses the British
2:09:33pound and then everybody else uses the
2:09:35Euro
2:09:36right so I see why this is important is
2:09:39there not like one for uh China China's
2:09:43dollar and stuff whatever that
2:09:46is is not important or something
2:09:51Jane uses the British pound everybody
2:09:54else in Europe uses the Euro Japan uses
2:09:58the JPY I feel like the two missing from
2:10:01an outsider standpoint are the Yen yeah
2:10:05the Yen and then
2:10:08also
2:10:10India and
2:10:13Russia those are all the big ones I
2:10:16don't know if those are tradable though
2:10:18so it's
2:10:20interesting Swiss
2:10:22frank
2:10:25okay dang
2:10:27so yeah I'm not going to go down that
2:10:29that rabbit hole because I guess there's
2:10:31a lot more not everybody uses the Euro I
2:10:34was thinking everybody uses the Euro but
2:10:36I guess not like everybody on over
2:10:40there over there where y'all
2:10:43at everybody over there uses
2:10:46EUR Yan yen is Japan oh yen is
2:10:52you w you w anyways let's go ahead and
2:10:55analyze this data because we got a new
2:10:56model here and this is what it looks
2:11:00like it looks good to me did you see it
2:11:05look at
2:11:06this look at me 98%
2:11:17prediction the thing is the whole
2:11:19system's all dollarized so the majority
2:11:22of the flow is dependent on the US the
2:11:24Swiss frank is interesting because of
2:11:26how tied into International banking it
2:11:29is wonder what info could be inferred
2:11:32from analyzing
2:11:33that dang UJ is going to fly higher in
2:11:39this coming week what is UJ is that Euro
2:11:43or us
2:11:45Japan dxy is a good indicator of USD
2:11:49strength I think I got that right here
2:11:52yep
2:11:53dxy so dollar
2:11:57down risk assets
2:12:00up dollar up risk assets down that's
2:12:04what I learned at least I don't know if
2:12:07that's
2:12:08right but dollar down today and risk
2:12:11assets
2:12:13down so down BTC down interesting this
2:12:18game is super fun because there's so
2:12:19many different levels to it in different
2:12:21Arenas like y'all are over here in the
2:12:24uh Forex Arena and I haven't touched
2:12:26that I read baby Pips when I was in
2:12:29college maybe or whatever and I I was
2:12:32bored to tears I was bored to tears dude
2:12:35I was bored to tears that baby Pips
2:12:38there's so much information there but I
2:12:41probably could I probably could dive
2:12:43back in now that I'm a little more
2:12:44financial Savvy I
2:12:47guess that's my interpretation too
2:12:49Connor says
2:12:51it is half right it's half right until
2:12:54like a big down day like this I feel
2:12:57like then it's like everything just hits
2:12:59the fan that's what I've observed at
2:13:02least is that like on days that it's
2:13:06slightly
2:13:08down risk assets up if it's mad down
2:13:12then everything down dxy is a basket of
2:13:15currencies against the
2:13:18dollar so if people are in Europe and
2:13:20want to invest in US Securities they
2:13:22need to raise the US dollar
2:13:26interesting well let's stop let's stop
2:13:28talking about that stuff because I I
2:13:30don't know where that's going to take
2:13:32me so maybe in those days there's a
2:13:35rotation from Euro to USD dang it's
2:13:38awesome there's so many games in this
2:13:40game that's why like I'm not worried
2:13:42about allocating time to algorithmic
2:13:44trading like you you allocate time to
2:13:47like new trends like Shopify or apps or
2:13:52I don't know social media you don't know
2:13:54if it's going to be here in 5
2:13:57years I needed something that was so
2:13:59hard and I know that's going to be here
2:14:02for the next 60 years and I know for a
2:14:04fact there's going to be some sort of
2:14:06Market here for the next 60 years so
2:14:08every single minute I put into this
2:14:10stuff I'm
2:14:12compounding and I know in the future
2:14:14it's going to be
2:14:15useful I'm getting better and better
2:14:17every day and thankful to you guys that
2:14:21like accelerate this stuff too so if
2:14:23you're in the chat you see everything I
2:14:25see I wouldn't miss it I wouldn't miss
2:14:28it if I were
2:14:30you thank you for providing the results
2:14:33of both models I'll compare them based
2:14:36on the given metrics and features State
2:14:39prediction accuracy new model adx ATR
2:14:42and Donan 98% model
2:14:452 volume change BB width volatility 89
2:14:50the new model model shows higher
2:14:52accuracy in predicting the next
2:14:56state new
2:14:58model log
2:15:01likelihood minus
2:15:0693,000 model 2 plus
2:15:1114,000 model 2 has a much higher log
2:15:14likelihood indicating a better
2:15:17fit back to model 2 Bic basian
2:15:20information
2:15:22Criterion damn
2:15:24187,000 verse negative
2:15:2827,000 lower Bic indicates a better
2:15:31model model 2 has a significantly lower
2:15:34negative Bic suggesting it's a better
2:15:36fit cross validation score new
2:15:41model
2:15:42-47 47k verse 20 2000 model 2 has a
2:15:47higher and positive cross validation
2:15:49score indicating better generaliz ation
2:15:51to unseen data import feature importance
2:15:55new model adx 27
2:16:0085 ATR
2:16:03point4 I like how it's a little more
2:16:08spread out but model 2 is all about
2:16:11volume essentially visual
2:16:13inspection the new model's chart shows
2:16:15more distinct and prolong State periods
2:16:18especially in the later half of the time
2:16:20series this could indicate that it's
2:16:22capturing longer term Market regimes
2:16:25conclusion while the new model adx ATR
2:16:27Donan shows higher state prediction
2:16:30accuracy and more balanced feature
2:16:31importance model 2 volume change BB
2:16:34width and volatility outperforms it in
2:16:37terms of log likelihood Bic and cross
2:16:39validation score this suggests that
2:16:41model 2 is likely the better model
2:16:45overall as it fits the data and
2:16:47generalizes more effectively to unseen
2:16:50data
2:16:51the heavy Reliance on volume change in
2:16:53model 2 appears to be capturing
2:16:55something fundamental about bitcoin
2:16:57price movements that others don't
2:17:00however the new models more balanced
2:17:02feature importance and distinct State
2:17:04visualization might offer different
2:17:09insights using model 2 is a primary
2:17:14model keeping the new model as a
2:17:16complimentary tool for additional
2:17:19insights all right this is dope dude I
2:17:22love this
2:17:24game cuz you know what's next a new
2:17:27idea and then you know what's next after
2:17:30that a new idea I don't care if I miss
2:17:32the shot bro I will just keep shooting I
2:17:34will keep shooting I know I'm a shooter
2:17:36I'mma keep shooting I'mma keep shooting
2:17:39dude I'm just going to keep shooting and
2:17:42keep shooting and then I'mma shoot again
2:17:45and then I'm going to shoot again and I
2:17:47don't care how much time this stuff
2:17:49takes dog cuz I'm here for 60
2:17:5160 years
2:17:54B I don't care I'm here
2:18:00present do you use volume as a
2:18:02percentage change yes
2:18:04sir volume
2:18:08change all right let's try another
2:18:18one use these three new
2:18:25features instead of the old ones in the
2:18:31attached
2:18:34code they are all from
2:18:38pandas
2:18:44ta yo I haven't watched pton fo probably
2:18:47for 20 years
2:18:48though so I don't know what I'm even
2:18:51referring to paid and full when did this
2:18:54launch yeah 20 2002 probably watched it
2:18:56when it came out or around that time I
2:18:59was a kid I need to watch this again as
2:19:01an
2:19:02adult because you know all that stuff
2:19:05goes over your
2:19:06head all of it goes over your head so uh
2:19:09I started watching this show I grew up
2:19:11on like uh fresh prince of belir I love
2:19:13that show yo they have a new show oh
2:19:16because you said B remind oh okay they
2:19:19maybe that's where I got it from
2:19:21B I don't
2:19:25know but
2:19:30um I don't watch TV too much but I start
2:19:33watching the new Fresh Prince it's
2:19:35pretty it's pretty good A lot of like a
2:19:38lot of references to old black movies
2:19:40which is
2:19:41dope
2:19:46Kan do you care if you live or you die
2:19:51that threw me off I was like I know what
2:19:53he's about to say that's exactly what
2:19:55Kane's grandpa said to him and they just
2:19:57LED this they started the show the new
2:19:59fresh PR show oh it's called belir sorry
2:20:03it's called Bel Air um it started with
2:20:07will you care if you live or you die and
2:20:10that's like you got to be you got to be
2:20:12watching all the old movies in order to
2:20:14understand the show fully which is dope
2:20:16I love it I love it I love it I love it
2:20:19and they give I love how they ALS like
2:20:21totally changed the show like Carlton is
2:20:27somewhat
2:20:28cool um
2:20:31jazz jazz finally swoops his
2:20:35girl whatever her name
2:20:39is
2:20:40uh Jeffrey is like an assassin L key
2:20:45he'd be holding it
2:20:48down it's crazy I love how they just
2:20:50like switched it up a little bit at
2:20:51first I was thrown off like why is Uncle
2:20:54Phil some like skinny dude rest in peace
2:20:58Uncle Phil I like how they did that at
2:21:00first I was like apprehensive towards it
2:21:02like why is Uncle Phil like is he he who
2:21:05is this guy but he's still a great
2:21:08father figure takes care of will I'm
2:21:10only on like the seventh episode so I
2:21:13was happy to see that there's three
2:21:15seasons to
2:21:16it it's a good show and I don't say that
2:21:19about many shows
2:21:21I I don't usually like watch this
2:21:24St a got time B yo I need to watch this
2:21:28movie again though got Regina a young
2:21:31Regina Hall we got Cameron in it hey Ma
2:21:35that used to be my
2:21:37head I used to sing that all day
2:21:43long damn there's some good movies I'm a
2:21:46old head dude there just some old head
2:21:48stuff to be saying back in my day back
2:21:51in my day we had some good
2:21:54movies cast is Elite absolutely damn I
2:21:58can't wait I might watch this tonight I
2:22:00mean I'm sure it's on somewhere
2:22:03Netflix eight more where to
2:22:09watch
2:22:12YouTube I got Amazon
2:22:15Prime thank
2:22:18you you're an territory I know it's uh
2:22:22it's crazy dog it's really crazy I mean
2:22:25what it is what it is but I am I'm there
2:22:28I'm
2:22:29there off officially an
2:22:32unk you never think you're going to get
2:22:34here till you get here dude I'm
2:22:37literally an
2:22:38uncle I got two two little Young
2:22:43Bucks and it's cool it's cool man it's
2:22:45cool to see cool to see cool to see him
2:22:49jack my SCH SW my swag is irrelevant but
2:22:52they Jack it anyways because I'm you
2:22:55know so somebody got to give it to
2:22:58him all right so I'm writing down Pon
2:23:01full real quick all right let's get back
2:23:03to it I'm going to go ahead and test one
2:23:06more so use these three new features
2:23:09instead of the old ones in the attached
2:23:11code Lin regression macd and true range
2:23:15see I like just switching it up entirely
2:23:18so our best one is volume change
2:23:25but we just tried the adx and it at
2:23:28first it looked better because of the
2:23:3398% it looked better but was it better
2:23:37no
2:23:39way uh send me back send me back all the
2:23:44code all right I love this game dog we
2:23:47literally are competing with the the
2:23:49machine learning people over
2:23:52at name your Fang name your Fang you
2:23:56gave us too much power dude you gave us
2:23:58too much power Claudia you gave us too
2:24:00much power Gigi you gave us too much
2:24:03power
2:24:05luxie you don't understand it okay go
2:24:08ask
2:24:15AI I say CO's the great equalizer but
2:24:17like damn AI is really the great
2:24:19equalizer but you got you got to know
2:24:21how to code you got to know how to code
2:24:23to be able to get the best out of AI
2:24:24because what are you using it for
2:24:26otherwise right like emails and stuff
2:24:29okay I see through I I can see through
2:24:31that you can't see through my code you
2:24:33know what I'm saying like code is code
2:24:35so like you can kind of feel when AI is
2:24:37talking to you almost I mean not always
2:24:40but like how often are you writing
2:24:42papers how often are you writing emails
2:24:44I mean how often are you writing
2:24:46Facebook posts I mean it's cool whatever
2:24:48like that's a cool use case for it but
2:24:50coding is the killer app coding is the
2:24:54killer app for AI right now and if you
2:24:57don't know how to code you can't use it
2:24:58properly I mean you can learn through it
2:25:01at least but you got to you got to be
2:25:02able to learn it so that sends me to
2:25:06let's let's let's actually go through it
2:25:08because I know there are people here
2:25:10that weren't here at the start of
2:25:13this so this one's Lin Rag macd and
2:25:19TR l ly reg Mac and
2:25:24tr. py paste it in and then we're going
2:25:27to just do a little study session here
2:25:28because I need to see it dude like I
2:25:31said I I grew up getting 500 shots a day
2:25:34dog and this is the same game this is
2:25:36literally the same game you get better
2:25:37over time get wetter over
2:25:41time every single day I don't care how
2:25:45tired I am I don't care don't care how
2:25:47little I slept last night I don't care
2:25:49that's when I really double down and I
2:25:51make sure to get my hours in so import
2:25:53these are a bunch of imports then we're
2:25:55starting the hmm analysis on BTC we're
2:25:57loading in pre-processing we're dropping
2:25:59a column because my data is janky we're
2:26:02creating the index okay open high low
2:26:04close this is the linear linear
2:26:07regression with the TA pandas ta Library
2:26:10pandas ta Library so linear regression
2:26:13is a type of machine learning as well so
2:26:15cool we got macd here calculating the
2:26:18macd we're getting the true range here
2:26:20then we're dropping any na any non I'm
2:26:23actually about to go get some mother
2:26:24naan right now
2:26:29be train hmm seven components lyre macd
2:26:35true range I swear there's a new na
2:26:37place I mean it's Indian place but you
2:26:39know you know how you know how that goes
2:26:41train hmm we've got seven components
2:26:44that's seven different regimes we are
2:26:46training on but we have features three
2:26:48features linear regression magd and true
2:26:52range we normalize those features
2:26:54meaning put them between zero and one I
2:26:56believe then we run the hmm the goian
2:27:00hmm are there other hmm that we can
2:27:06run put it on the read me put it on the
2:27:09read
2:27:14me we are using the
2:27:18goian hmm are there
2:27:22others that we can
2:27:26run where we at where we at ly red okay
2:27:29so and now we the Asian is completed we
2:27:33predict States so it's predicting States
2:27:41remember this is the actual prediction
2:27:45here we have seven different states but
2:27:47the thing is is it doesn't actually know
2:27:49the states it's just ch them in the
2:27:50states and then we're going to have to
2:27:51do that later but that's okay I just
2:27:53want to see a lot I want to see a lot
2:27:55and then see what the best is of a lot
2:27:57since I'm starting at a zero point
2:27:58origin where moev knows nothing I want
2:28:00to see a lot of different tests here and
2:28:02then out of those tests imagine you've
2:28:05never played a game before I have no
2:28:07idea how to play this game I'm going to
2:28:09play the game a 100 times and see my
2:28:11score over the course of a hundred times
2:28:14and then I'm going to go with the method
2:28:16of the best score that's all I'm doing
2:28:19that's that's all I'm ever doing that's
2:28:22what this whole this whole world is
2:28:23about to me right
2:28:26now that's what life is about really
2:28:29just a test your way into figuring
2:28:31things out predict next state save
2:28:33changes it's just saving the
2:28:35changes St
2:28:38changes pending it to a
2:28:41folder and we haven't even been looking
2:28:43at them but
2:28:47okay model scaler where are those
2:29:02I can delete these actually need to
2:29:05update them real quick I'm going to
2:29:06update them in my thing so you can see
2:29:09here that I updated the file path so
2:29:12here at the
2:29:16bottom I added a folder to all these so
2:29:20I got to just update this real quick um
2:29:23hmm model 2 here every time I run this
2:29:26it's going to try to save it again
2:29:29so this is our best model so far model
2:29:33one here second best actually I don't
2:29:36know if it's second best but I don't
2:29:37really care about the second
2:29:38best I just want the best be that's it
2:29:43just trying to be the best I can be I
2:29:44don't know how good I can be but been
2:29:47able to figure it out thus far so I'm G
2:29:49keep going
2:29:57plot the results okay that's just a
2:29:58bunch of plotting I just want to make
2:29:59sure you see all this code so you can
2:30:01leave here with your hands
2:30:03full of ideas and
2:30:07code data saved to a
2:30:10plot calculate prediction
2:30:14accuracy calculate Bic Bic is a
2:30:17evaluation technique so something about
2:30:20basy and something something something
2:30:22death time series CV okay analyze
2:30:25feature importance okay starting main
2:30:30execution training the model predicting
2:30:33States these are the names of the states
2:30:35but remember these names are just made
2:30:36up
2:30:38really the model we have to place the
2:30:41states later this is just running all
2:30:43the code but I just want to make sure
2:30:44you see it all try to show you all the
2:30:46code but sometimes I'll be forgetting
2:30:48dog that's my my bad Mal M
2:30:53bro Lyn reg 14 key error come on cousin
2:30:59I showed you all the wrong code then
2:31:01it's not done yet calculating
2:31:05lyre okay I get this
2:31:10air but I can see in the available
2:31:16available
2:31:18indicators that
2:31:21let's see what the available indicators
2:31:22look like these are all the available
2:31:24indicators on Lin
2:31:29rag I have these two
2:31:33options I have uh
2:31:37indicators on Panda ta that I have these
2:31:42options
2:31:46here okay um let's see they gave us too
2:31:50much power llord said I think they did
2:31:52and they can't take it back now was the
2:31:53crazy thing you can't go backwards with
2:31:55this they did give us too much power
2:31:57they gave us kids out here too much
2:31:59power now noobs you can't tell that I'm
2:32:01a Noob anymore like if you look at my
2:32:03code if you look at my code you can't
2:32:06tell that I'm I'm just at the start of
2:32:08the game I'm only at my 3400 hour and I
2:32:12have to get to 10,000 hours before I
2:32:14know anything before I'll ever people
2:32:17ask oh moev get on my podcast tell me
2:32:19this and it's like bro I don't know
2:32:21nothing yet I'm not even halfway there
2:32:24yet stop it maybe I know more than some
2:32:28but 4,000 hours no way I need to get
2:32:3110,000
2:32:33hours I need to get to 10,000 hours
2:32:35before I can confidently say I know
2:32:38anything I'm sorry dude sorry to do it
2:32:42to you I'm sorry I'm not your Guru sorry
2:32:45I don't know everything in the world
2:32:50the only thing I do know is I will keep
2:32:52going I will keep finding new Secrets
2:32:55live every single day there's aha
2:32:58moments every single day
2:33:01dog every single
2:33:04day take care make sure you're
2:33:09here I think uh man man Manuel
2:33:14maybe I think learning writing and
2:33:16coding separately are very important
2:33:18like me I'm learning the English
2:33:20language because of AI
2:33:23sick that's Fire Dog you're born in New
2:33:27England so you're learning English
2:33:29because of AI That's fire okay I could
2:33:32see that I can see that learning just in
2:33:34general with AI is crazy like anything
2:33:37you want to learn it knows mostly
2:33:38everything at leasts a structure to
2:33:40everything that's awesome you're learn
2:33:42English and
2:33:45coding this would be killer for resumés
2:33:47too absolutely dude the bay indicator
2:33:51thing is a comparison of the log
2:33:54likelihoods it's used to gauge whether
2:33:57you could choose one model or the other
2:34:00thank you bro so the Bic it compares the
2:34:04log likelihoods which we've seen a lot
2:34:07of and we're about to see it
2:34:10again Bas factor basian basian factor
2:34:14base
2:34:15Factor
2:34:18so that's cool man you show up to these
2:34:20lives dude you learn at exponential Pace
2:34:22because I'm pushing us into a direction
2:34:25that I I'm interested in every day and
2:34:29then there's people way smarter than me
2:34:31up in the chat that are so gracious with
2:34:33their time and gracious with their
2:34:35thoughts and they went to school for
2:34:37this stuff dog I ain't go to school for
2:34:44this let's try it again
2:34:46doggy doggy style
2:34:52sad sad did I not put the right code in
2:34:55h you
2:34:56nerd I apologize for the confusion
2:34:58you're right and thank you for providing
2:34:59the information the error occurs in the
2:35:01Lin R function in pandis ta with
2:35:04different column names okay let's modify
2:35:06it so we need to put the load and
2:35:08pre-processed data we need to change
2:35:14that here are the other parts of the
2:35:16code that need to be updated okay I'll
2:35:17do that all right do
2:35:24it load and pre-process data
2:35:47here beaters
2:35:50yo what up Nick how are
2:35:54you how are you Nick have never seen
2:35:57your name before so welcome welcome
2:35:59welcome welcome in the predict States
2:36:04function so I'm going to go to all the
2:36:06Lin Rags
2:36:08here train hmm
2:36:13here what did it change do I don't think
2:36:16it changed
2:36:18anything anything think is plenty
2:36:30man that looks guzy to me
2:36:34beef all right let's just run it see
2:36:37what happens
2:36:54send back full code with any prints you
2:36:59need to debug my dog
2:37:31does unknown says I upgraded Twitter to
2:37:34get grock but can't handle heavy coding
2:37:39challenges wonder
2:37:42why PO is the best amount for free
2:37:47interesting Nick says he a newbie here
2:37:50salute salute to you dude much love 777
2:37:54how pass Love Good Vibes all the time
2:37:57you'll see it all the time in the chat
2:37:59light the chat up with the 777s for our
2:38:01new friend
2:38:03Nick so much love so much love all right
2:38:06let's go ahead here and try this new
2:38:08code out see what you got Litt
2:38:11Mama
2:38:13nothing linear regression calculating
2:38:16time series dang we're stuck this is
2:38:19good this is good exercise
2:38:24though good
2:38:29exercise Jane says 777 welcome to the
2:38:32fan fam
2:38:33fam wow dollar huh look at that dollar
2:38:37we look at the dollar now BTC this
2:38:41liquidation Point still got me tripping
2:38:44here 150 mil liquidated right here going
2:38:48downwards came all all the way back up
2:38:50did a cherry kiss to
2:38:53it and then reverted right back down and
2:38:57then jump got liquidated right
2:39:02here that's wild that's wild that's all
2:39:05I've got to say though because I don't
2:39:07want to predict Market movements or
2:39:10anything like that here it's not what we
2:39:12do here
2:39:13dude we let the AI do the
2:39:16work yo I like your pfp bro
2:39:21Nick it's a good
2:39:23one yo what happened to the live stream
2:39:26yesterday I cut it off randomly didn't
2:39:28I I forgot I had a hard stop at 3 and
2:39:31then it was
2:39:34305 West
2:39:43Side see what she found out here
2:39:51I apologize for the confusion it seems
2:39:53that ta Lin red function is returning a
2:39:55series instead of a data
2:40:17frame run it back then
2:40:21there we go
2:40:22shorty we can do it all dog we can do it
2:40:25all where's the m d at
2:40:44though okay there we go beautiful
2:40:46looking chart I'll tell you that much
2:40:53beautiful all right let's analyze it do
2:40:56exactly what we did with the last one
2:40:59repetition is key dude for me and this
2:41:02is my world I'm sorry if you're way
2:41:05ahead of me and this is too easy for you
2:41:07but for me it's a challenge I like I
2:41:10like challenges but I got to understand
2:41:12this stuff fly FY because I got to be
2:41:16able to grab it by the mother balls and
2:41:18then just understand all okay you know
2:41:21if I understand it all I'm a threat if I
2:41:23don't understand it all I'm not a
2:41:26threat dude I'm just following in the
2:41:29footsteps of Jim Simons
2:41:31he he did this till he died
2:41:35dog that's
2:41:38amazing it's amazing I'mma do it till I
2:41:41die
2:41:45too I'm do it till I die
2:41:57I'm just waiting on this stat these
2:41:59stats to be done
2:42:00here and I actually think I have my um
2:42:04best one here show all
2:42:08indicators the if I got to read me
2:42:11here I wonder why it takes a little bit
2:42:13of time in between
2:42:16this okay here we go
2:42:23let's see if she can just compare them
2:42:26to my model
2:42:31to to the previous ones let's just not
2:42:34even give it to
2:42:36her let's see what she really about
2:42:39let's see what Claudia is really
2:42:44about see if she can really memorize
2:42:46things nine more messages get off my D
2:42:48you know I'm running it up every single
2:42:50day nine more messages why do I even pay
2:42:53you fool what you even pay
2:43:02for who even pays for this stuff
2:43:10dog I
2:43:13do dag
2:43:16NAIT Jiminy Christmas certainly I'll
2:43:20compare the results of this new model
2:43:21using Lin Rag macd and true range with
2:43:25the previous two models we discussed
2:43:27let's break it down by each metric and
2:43:29let's get that thing back up
2:43:31here let's get the image the
2:43:35image
2:43:37open revealing finder please plot it
2:43:42plot It Like It's
2:43:45Hot Leg is hot baby
2:43:49my leg is
2:43:54hot yeah you don't get much I mean you
2:43:56get a good amount
2:43:59Jane but you can see I only got eight
2:44:02more messages
2:44:04left for like three more hours this
2:44:08crazy stuff so I think I'm going pretty
2:44:12hard on it today
2:44:14but um I feel like it's less than
2:44:17yesterday I'll say that much
2:44:22feel like it's less than
2:44:29yesterday I've been
2:44:31using kote says I've been using open AI
2:44:34API plus pine cone DB to avoid the rate
2:44:40limit okay okay I'll feel you what is
2:44:44pine cone DB
2:44:53oh the V vector vector B I remember this
2:44:56tripo yo I got luxie so I'm Gucci like I
2:44:59got a backup I got luxie is you know
2:45:03unlimited
2:45:06so she's my
2:45:08backup and she does great it's using all
2:45:11GPT stuff so I'm Gucci on that on the
2:45:15backup so I was just complaining to
2:45:16complain cuz but I appreciate all the
2:45:19tips because this is how how we find out
2:45:22about new stuff you know
2:45:56okay my monthly spends looking pretty
2:45:58solid
2:45:59here on
2:46:01gbt I was trying to keep it down a
2:46:03little bit but you know it's all good
2:46:06we're going to use luxie here when we
2:46:08get worn out of or we're done with the
2:46:12whatevers so here's the model here that
2:46:16we just
2:46:17tested and this model uses adx the ATR
2:46:22no sorry new model is linear
2:46:27regression macd and true range
2:46:3297% and we're comparing it to our other
2:46:34model or other two models now previous
2:46:37model was
2:46:41098 so that's better there model two
2:46:44though which is our best model has a
2:46:47lower prediction
2:46:49State prediction accuracy the new model
2:46:51performs very well in terms of State
2:46:53prediction accuracy almost matching the
2:46:55adx ATR daan model and outperforming the
2:46:59volume change model new model is 30,000
2:47:02minus 30,000 and the other one was minus
2:47:0593 so it made an improvement model 2 is
2:47:0814,000 the new model shows a significant
2:47:10improvement over the adx ATR but still
2:47:13doesn't reach the positive log
2:47:16likelihood of volume change that's funny
2:47:17our first one of our first models or
2:47:19second model Bic Bay in information
2:47:22Criterion new model
2:47:2461k previous model was 187k and then
2:47:28model 2 was minus 27k the new model has
2:47:31a much lower which is better Bic than
2:47:35the adx ATR Donan model indicating a
2:47:38better fit while accounting for model
2:47:41complexity however the volume change
2:47:43model still has the best lowest Bic
2:47:47cross validation score new model is-
2:47:52100K previous model is -
2:47:5550k and I think this is the important
2:47:57part the standard deviation model 2 is
2:48:002000 still the best the new model shows
2:48:04worst cross validation performance
2:48:06compared to new previous model with a
2:48:08lower mean score feature importance new
2:48:11model Lin regression um 47 so this
2:48:15doesn't matter too much it shows more
2:48:18balance which is nice visual inspection
2:48:21the new models chart shows more frequent
2:48:23State changes compared to the previous
2:48:26ones particularly in the earlier part of
2:48:29the time series this could indicate that
2:48:31it's more sensitive to short-term price
2:48:33movements conclusion the new model
2:48:35linreg macd true range shows
2:48:38improvements in some areas particularly
2:48:40in state prediction accuracy and log
2:48:43likelihood compared to the adx ATR Donan
2:48:46model however it still doesn't
2:48:48outperform the volume
2:48:50change the new model's feature
2:48:53importance is more balanced than the
2:48:54volume change one which might provide
2:48:56more diverse insights into Market
2:49:01Behavior the frequent State changes in
2:49:04the visual representation suggests that
2:49:06this model might be more reactive to
2:49:08short-term price movements overall while
2:49:11the new model shows some promising
2:49:13characteristics especially in its
2:49:15balanced feature importance and high
2:49:17State prediction accuracy it doesn't
2:49:19conclusively outperform the volume
2:49:21change model across all metrics the
2:49:24volume change model still appears to be
2:49:26the most effective in capturing the
2:49:28underlying dynamics of Bitcoin price
2:49:32movements okay sick well let's keep it
2:49:34moving then let's try it again I'm going
2:49:37to take all this
2:49:47information just plop it in there
2:49:49because why wouldn't
2:49:51I just plop her
2:49:57in plopped it
2:50:11in perfect perfect I'm getting
2:50:13comfortable with this I can see ly
2:50:15regres macd true range like sometimes
2:50:17this stuff's so intim then you can't
2:50:19even see
2:50:20it you know you can't even look at it
2:50:23like don't even look at me don't yo you
2:50:25are so complicated I don't even want you
2:50:27to look at me right now CL no don't look
2:50:30at me stop looking at me dude could you
2:50:33imagine if I was on camera there's no
2:50:35way the show would not go on if y'all
2:50:38fools were looking at me this whole time
2:50:40I would stop it stop it stop
2:50:43it stop it stop it stop it y'all would
2:50:45fall in love with me you already fall in
2:50:48love with my voice
2:50:52voice 97% males in the audience by the
2:50:57way
2:51:01weirdos oh my
2:51:04god let's go ahead and try some
2:51:08more so what I'm going to do do here is
2:51:11I'mma paste this
2:51:14in the things I do without actually
2:51:17thinking I did is crazy you see that
2:51:20over and over again like moev you
2:51:23already did
2:51:24that look at this look at this
2:51:27professional looking file
2:51:31dude this is a good looking I'm GNA push
2:51:33this to GitHub right now this is just a
2:51:36good looking non moev
2:51:40like dang really cool really cool I
2:51:44remember back in the day when I saw
2:51:46folders I would just oh pretty much just
2:51:50puke I pretty much just puke because
2:51:54it's so intimidating looking at this
2:51:56code over here and then folders of the
2:51:58code how do you
2:52:00even oh man that's why it's hard to work
2:52:03on other people's code bases in my
2:52:05opinion that's one of the reasons
2:52:24what is this fer called hidden Markoff
2:52:27model rest in peace Jim so much love to
2:52:30you dog you don't understand how
2:52:32inspiring you are
2:52:34dog appreciate every every thought you
2:52:37put out every thought you put out in the
2:52:40universe I know you're not going to show
2:52:41us everything and now we'll never know
2:52:43but I feel like you gave us enough
2:52:46enough pieces I'm grateful I'm grateful
2:52:49for him thank you Jim rest in peace 777
2:52:52to you your loved
2:52:54ones I hope there's somewhere amazing
2:52:57that you had it after
2:53:00death and other gy too I love you like a
2:53:03pops to
2:53:05me much love Eerie that you both died in
2:53:08the same year Eerie Eerie stuff when I
2:53:12say Jim I'm really talking to both
2:53:16y'all much love
2:53:19pops let's go ahead here
2:53:29and let's try these new
2:53:33ones
2:53:35so instead of using the current
2:53:40indicators in the above
2:53:43code use these three
2:53:48from
2:53:50ta all right and one of them is a
2:53:53Candlestick that's weird I don't know if
2:53:56it'll work
2:54:02but
2:54:08tripo metac CLA GPT did you check grock
2:54:13NOP
2:54:15nope so the vector database I remember
2:54:17that from uh um Auto GPT
2:54:22days Jane says grock many
2:54:25apis it looks like it's a drop in code
2:54:28for open AI compatible so not much
2:54:33changes that's
2:54:35dope feel like I got my trippy kit for
2:54:38now but keep it coming keep it
2:54:46coming so I need to go to H mm models
2:54:49what was the last one we did we did Lin
2:54:53rag so I'm going to copy all this code
2:54:56over and this new one is
2:55:00called Stoke comma
2:55:10Hammer yo am I doing this right like
2:55:13that's my question is this what is this
2:55:16what machine learning is it's just
2:55:17coming up with a bunch of ideas and
2:55:18testing them anybody in the audience
2:55:21know
2:55:23like you know are you supposed to switch
2:55:26models more often am I doing this right
2:55:29yeah I guess that's the question I feel
2:55:30like I'm just going in a loop which is
2:55:33good because I'm seeing new data but
2:55:35again I'm trying to get like 50 shots up
2:55:37just so I can see where to go but I'm
2:55:39curious if there's another approach to
2:55:41this I'm actually just GNA ask
2:55:44AI I'm running out of I'm running out of
2:55:47chat so I'm not gonna I'm not going to
2:55:48ask her not right now but I'll put it on
2:55:51the Note sheet let's go to read
2:55:57me ask AI am I doing this
2:56:02right am I
2:56:05literally
2:56:08supposed to
2:56:11just just test all of these different
2:56:17indicators
2:56:20indicators and
2:56:23ideas with different models and
2:56:30features to predict different
2:56:36things I know price isn't the best to
2:56:41predict but test uh
2:56:45predicting Market r
2:56:49regimes
2:56:51and when to run
2:56:55each
2:56:57um
2:56:59strategy could be good ideas but just
2:57:05curious if this is the right
2:57:09route
2:57:11route to take When
2:57:15approaching ml for trading
2:57:19is it really just about testing all
2:57:24these different
2:57:26ideas
2:57:29features
2:57:32indicators
2:57:35Etc trying to find a model that works
2:57:40best what is best I still I'm not clear
2:57:43on that my best right now is just I'm
2:57:45trying to compare it against others you
2:57:47know a different other ones that I do
2:57:50but you
2:57:58know Rick RI what's up
2:58:01bro good to have you here dude Anga G
2:58:05says you're just fine-tuning the
2:58:07parameters and putting the model through
2:58:09various tests you have designed to get
2:58:12it to reflect reality as close as
2:58:14possible
2:58:16essentially so do you feel like this is
2:58:18a good path I mean I feel like I haven't
2:58:20found anything better than the volume
2:58:22one yet and I'm going to keep going
2:58:26obviously keep going but I just want to
2:58:28make sure like when I come back at
2:58:30it that um it's the right path to take
2:58:35but I I like this so far I also want to
2:58:37try some should I be trying should I be
2:58:41trying rnn's or other um
2:58:46models where can can I find all the
2:58:50suitable models to use on this time
2:58:55series
2:58:58data do you have a GitHub you says Nah I
2:59:01mean I got a GitHub but it's a private
2:59:05GitHub
2:59:07um Kieran says first time here and I
2:59:09don't understand a thing on the screen
2:59:11LOL that's fun man or lady yo B I'm
2:59:16going stick it to B that's fun I'm I I'm
2:59:19really proud to hear that honestly
2:59:21because I didn't understand any of this
2:59:23when I got started either and it just
2:59:24takes like persistence day after day
2:59:29um I believe code's a great equalizer if
2:59:32you don't know how to code then you got
2:59:34to learn bro you got to
2:59:37learn that's the first
2:59:41step cuz then you can build
2:59:43anything riss says I'm pursuing a data
2:59:46science degree at the moment and have
2:59:48beginner coding experience what concepts
2:59:49do you recommend I study needed to make
2:59:52a trading
2:59:53bot damn you're going to school for
2:59:56this you should be recommending stuff to
2:59:58me
3:00:00honestly no but in all seriousness I've
3:00:02been coding here for 3 and a half years
3:00:04so I don't know where you're at but I
3:00:06show it all on my YouTube channel I've
3:00:09been tra building trading Bots pretty
3:00:10much
3:00:11specifically for three and a half years
3:00:14on my YouTube so you can just start here
3:00:16and just kind of pick through however
3:00:17you want
3:00:18I know you're in school so you probably
3:00:20don't have a lot of cash but the boot
3:00:22camp I have a boot camp that walks you
3:00:23through all this stuff step by step you
3:00:25can read about it here and probably
3:00:27reverse engineer it if you don't want to
3:00:28join but $69 I know you're already
3:00:30spending a ton of money at school so no
3:00:33worries if you can't afford it um you
3:00:35can always just learn stuff at a little
3:00:37little slower speed on my YouTube
3:00:39because everything that I've learned
3:00:41I've done here Live on YouTube or
3:00:43somewhat live and um
3:00:48I just put the good stuff in the boot
3:00:50camp so my YouTube has everything and
3:00:54people always ask me to do like short
3:00:55concise videos but this helps motivate
3:00:58me to keep going when I'm I'm kind of
3:01:00like showing you over my shoulders it's
3:01:01like you guys are watching me do this
3:01:03entire thing every single day for the
3:01:05last multiple years and I can't stop I
3:01:10can't stop because we're a squad now you
3:01:12know yo kieran's from Singapore and you
3:01:16is from Singapore yo you y'all are on it
3:01:20your whole country is on it I never been
3:01:22to Malaysia but
3:01:26um I know Singapore like y'all are high
3:01:29tech over there angan G says I think
3:01:32you're on the right path Industries like
3:01:34Aerospace weapons and GPS have un or
3:01:37utilize hmm the most but finding the
3:01:40position of something in a Newtonian
3:01:44space is way easier than markets ah yeah
3:01:47makes sense but I still think it's worth
3:01:50looking at if you're stuck for examples
3:01:52ah it's good
3:01:55idea it's a good idea to look at that
3:01:57stuff over there in other spaces okay
3:02:02so I was just I wanted to ask AI this
3:02:05eventually but for now let's go ahead
3:02:07and compare the data or what were we
3:02:10doing oh I was doing a new one Stoke
3:02:14comma Stoke Cala
3:02:20stoked stoked Kamala is about to be our
3:02:24president first woman president first
3:02:27second second Brown president dude yo yo
3:02:31yo yo yo yo that's sick if we got two
3:02:34brown presidents in my lifetime I don't
3:02:36care what side you're on dog I literally
3:02:39couldn't care
3:02:41less I vote for race
3:02:48that's kind of whack but that's kind of
3:02:51whack
3:02:52but if you know you know you know and
3:02:56now gender I'm voting for a woman every
3:02:59time I can't vote but that's my input
3:03:03I'm voting for a woman every time way
3:03:05more
3:03:08organized and then you put you put the
3:03:10race card out there you know I'm voting
3:03:14Brown so
3:03:18you do you dude but you know my take on
3:03:22that I ain't going to vote anti Brown
3:03:25that's for
3:03:27sure that's pizy all right States my
3:03:32opinion doesn't matter though I'm not
3:03:33able to vote
3:03:35unfortunately or
3:03:38fortunately probably
3:03:42fortunately okay let's go ahead and
3:03:44paste those in there main execution
3:04:00okay past her in okay let's see what
3:04:03she's got let's see what Cala
3:04:06got I know I say her name wrong
3:04:09but I should I should correct
3:04:12that I should correct that
3:04:21kamla kamla didn't work on this one
3:04:24let's go ahead and see if what she can
3:04:26do come on baby
3:04:41girl there she go
3:04:48down bang
3:04:54bang your trading bot can automatically
3:04:57place orders on centralized exchanges
3:05:00yep um as long as you have an account
3:05:02you should be able to Via bance yep hey
3:05:06Mund dad you've inspired me to learn
3:05:07dude that's so
3:05:09dope I'm so happy for you
3:05:13bro learning coding is like the best
3:05:16decision in your life life in my
3:05:22opinion that is the best decision you
3:05:25could ever
3:05:29make so let's go ahead and copy the load
3:05:33and pre-process
3:05:41here and our
3:05:43our comma
3:05:46code comma code code see what our comma
3:05:49code does
3:05:51now oh but this might be a problem we'll
3:05:55see I've never used the Candlestick
3:06:00before what AI model you use uh so I'm
3:06:05using the hidden markof model right now
3:06:08just on day four though so I'm still
3:06:10barely new what type of models do you
3:06:12use for time series data we're trying to
3:06:14predict the state so not necessarily the
3:06:16price predict the uh regimes this is
3:06:19based off of some hints that um our guy
3:06:23Steve Jobs oops I mean Jim
3:06:27Simons okay
3:06:29dude same thing same
3:06:32thing I don't know I don't know dog
3:06:35Visionaries B
3:06:38Visionaries anyways we're using hidden
3:06:40Markov models because homeboy said he
3:06:42used
3:06:44them or his people used them but it's
3:06:48not only the only thing he used so any
3:06:51ideas you got please do drop them in the
3:06:53chesy predict States and analyze States
3:06:56here so this is a hidden Markoff model
3:06:59predict and analyze States we're trying
3:07:01to predict seven different states I
3:07:03think I'm going to start testing after
3:07:05this one with less States though because
3:07:09I think seven is
3:07:15aggressive use that it can help but in
3:07:17combination with something interesting
3:07:20so how would you combine it I
3:07:23guess and what like combin it with what
3:07:26like lstm
3:07:27[Music]
3:07:30or change oh did you take
3:07:37out she took it out
3:07:49why did you take it out of
3:07:52here look at that let's see what she
3:07:56said at least apologize for the error
3:07:58the issu is that likely caused by the
3:08:01nature of CDL indicator which returns
3:08:03discreet values usually minus 100 yo
3:08:06homeboy said this earlier dude listen to
3:08:08the chat the chat is where everything's
3:08:11at dog the chat is where everything's at
3:08:15somebody said this earlier and I just
3:08:16didn't understand it
3:08:18yet see that's the thing sometimes y'all
3:08:21say stuff I don't understand and I love
3:08:23that I love that because then I see it
3:08:26later and it's like okay now I
3:08:27understand what homeboy was trying to
3:08:28tell me
3:08:31earlier I love it it's a whole another
3:08:34element to this game and if you're here
3:08:35live you get the same you're here like
3:08:38you get the same things and you probably
3:08:39come to even better ideas cuz you see
3:08:42the chat
3:08:43too people leave with love around here
3:08:46this ain't scary scarcity set mindset
3:08:49Finance no way not around here I know
3:08:51there's a couple scarcity MOS up in the
3:08:53the chat right now but they ain't
3:08:55chatting they ain't chatting in the
3:08:56audience I say that there's some
3:08:59scarcely L mos in the audience but not
3:09:01in the chat chat chatly would love chat
3:09:04brings all ideas I apologize for the
3:09:06error the it's because the negative 100
3:09:09or zero or 100 doesn't continuously like
3:09:12other indicators this can lead to
3:09:13singularities Let's modify approach
3:09:16we'll keep stoke SI I and comma but
3:09:18replace CDL Hammer with A continuous
3:09:21indicator a good choice could be the
3:09:24average
3:09:25directional
3:09:28adx all right I'm with it I'm with it if
3:09:31you with
3:09:32it I'm with if you with it
3:09:48she sneaked me though she sneaked me
3:09:51real quick on that that's good though I
3:09:53love it I love it I love it I love it
3:09:55let's go push me to the limits
3:09:58dog push me to the edge all of my
3:10:02friends are
3:10:03dead push me to the
3:10:14edge all right
3:10:17P her
3:10:23in okay let's run
3:10:30it there we go so Vladimir what up bro
3:10:34I'm glad you're here dude never seen you
3:10:38before he says use the hmm in
3:10:41combination with something find state
3:10:44for example and then combine with other
3:10:47Model H boost or some
3:10:50signals
3:10:53interesting interesting stuff you got
3:10:56there dude I don't quite understand it
3:10:58but I will so yeah yeah let me just ask
3:11:02then if you don't
3:11:04mind let me ask you a question about
3:11:06that but the question would
3:11:09be wait a
3:11:12minute the question is okay so he says
3:11:15to use the hmm in combination with
3:11:17something like hxg boost okay
3:11:19comfortable with
3:11:21that find state what does that mean find
3:11:25state for example you mean like the
3:11:27regime like the states okay that's I've
3:11:29been working on for the last 3 hours and
3:11:31I don't even know what state means come
3:11:32on
3:11:34mondev come on
3:11:36mondev so we find the state and then
3:11:39combine it with another model XG boost
3:11:42or some signals so do you mean to
3:11:43predict price after so predict price in
3:11:47the states while predicting price is
3:11:49pretty Noob and everybody is trying to
3:11:51predict
3:11:52price is everybody trying to predict
3:11:55price during specific states there we go
3:11:58see now we're a level deeper see
3:12:01everybody wants to predict price I don't
3:12:02know is this right am I going there
3:12:04right Vlad I don't know everybody wants
3:12:06to predict price right
3:12:09away but if everybody predicts price the
3:12:11price ain't going to be the price
3:12:12anymore get it does that make sense if
3:12:15everybody predicts price and the price
3:12:16ain't going to be the price
3:12:20noo but not
3:12:23everybody he says few working on algo
3:12:27trading already few years
3:12:32dope dope dude but not
3:12:36everyone
3:12:37predicts
3:12:39regime and then predicts
3:12:45price the this is one level
3:12:50deeper is it the
3:12:54answer maybe not
3:12:57but but now we thinking really I like
3:13:10it I like it though
3:13:18I love it I love this game so grateful
3:13:20to
3:13:22play so so grateful dog that we have a
3:13:25long Runway here a unlimited Runway to
3:13:29play this game come on B let's go come
3:13:34on you do you whatever you do do it hard
3:13:37dog rest at the end not in the middle
3:13:41rest at the end come on be let's get it
3:13:44I'm talking to me dog I apologize for
3:13:47for people that don't like hearing that
3:13:48but it's me I'm going every day this is
3:13:52the day to
3:14:10do I mean we could back test things too
3:14:14oh my God bro stop it this game is too
3:14:17fun there's too many levels there's too
3:14:19many ideas
3:14:22dude we can back test some of this data
3:14:28too like buy when in
3:14:32state one and then sell when in state
3:14:39three
3:14:40unlimited ideas
3:14:43dog unlimited Come on B let's go
3:14:48go um let's see here vad says yes you
3:14:53find state for example one bullish
3:14:55signal plus then XG boost can predict
3:14:58your direction for example with 80%
3:15:01confidence if bot match you put long if
3:15:05both match you put long you're fire bro
3:15:08you are fire you are fire I like your
3:15:12mind dog I'm so happy that you're
3:15:15here I'm so so happy that you're here I
3:15:18hope I hope you come back jump in the
3:15:22Discord you're you're my people
3:15:25bro it's good to good to meet you I like
3:15:29how you think that's
3:15:31a bu
3:15:34idea buy idea I'm giving you the crown
3:15:36I'm giving you the crown for the day I'm
3:15:38sorry just you just hopped in dropped
3:15:40the mic and yeah yeah duh like of course
3:15:44of course you do it that way and now
3:15:46it's like cuz predicting price
3:15:48everybody's predicting price just put
3:15:51some put some get some separation in
3:15:53between predicting price and everybody
3:15:54else you know okay if you want to
3:15:56predict price everybody's got access to
3:15:58XG boost everybody's got access to open
3:16:00high low close volume
3:16:02data landlord said send details
3:16:08bro oh man you ain't got to send nothing
3:16:12you ain't got to send nothing more Vlad
3:16:14you drop mic you can you don't have to
3:16:16even come back honestly N I hope you do
3:16:19you're a homie and I hope you do
3:16:24but you provided your value thank you
3:16:27dog so much love
3:16:36C all
3:16:38right I said we going to be here for 60
3:16:41years
3:16:43I D there's unlimited things two tests
3:16:48all right sorry I'm just tripping now
3:16:50I'm just tripping relax relax I only got
3:16:53five more messages for today
3:16:57is I'm GLA I'm I'm over here glazing
3:17:00Vlad I'm over here glazing
3:17:04Vlad look at that I be learning hypers
3:17:07speed cuz of Y I learned a new I learned
3:17:09a whole new word today and then Vlad
3:17:12came up in here and dropped some dimes
3:17:18and then I use my new word because I am
3:17:21lowkey glazing right
3:17:26now oh man I love this game so grateful
3:17:28for this life that we live every single
3:17:30day we get to wake up and do anything
3:17:32that we want to do just close your eyes
3:17:34dude whatever you want just close your
3:17:36eyes 10 minutes a day dude I meditate
3:17:38for 1 hour a
3:17:41day I don't just say this stuff
3:17:44dog I really believe
3:17:48most people aren't thinking about what
3:17:49they want in
3:17:51life most people don't don't have the
3:17:53right direction they don't know which
3:17:55direction they want to go in they're
3:17:56just
3:18:00reacting but when you close your eyes
3:18:03and it's not just for sleeping when you
3:18:06close your eyes and you just sit there
3:18:07you can see where you want to go and you
3:18:08just go there it's going to be hard but
3:18:10it's supposed to be hard life supposed
3:18:12to be hard dog I love I'm so grateful
3:18:14for this let's get it let's keep going
3:18:17okay so here's the data analyze it so oh
3:18:22oh this one is poor I've never seen one
3:18:25so bad this is our worst one in the
3:18:28accuracy but let's see if it does
3:18:30anything else here's my here is the
3:18:35latest uh data
3:18:38output please compare to our other
3:18:42models
3:18:52so a Transformer model to predict price
3:18:55or Transformer model to yeah I guess
3:18:57that's that's the question
3:19:02right that's the question Transformer
3:19:05model to do
3:19:07what I've heard that a couple times now
3:19:10in the chat which is fire because when I
3:19:12see things a couple of
3:19:14times you know that people are
3:19:17people just leave with love around here
3:19:18dog this is the leave with love with
3:19:20Munda because I'm showing you literally
3:19:23everything I ever see every single day
3:19:25no matter what and y'all do the same
3:19:27thing back stop it I love it I love this
3:19:31world I love this world so so much you
3:19:34get what you put out dog you get what
3:19:35you put out all the time that's always
3:19:37you smile at somebody what do they do do
3:19:40they frown back no they don't frown back
3:19:45bro uh I found meditation very powerful
3:19:48and after was guided to Lord much closer
3:19:54dude don't look at accuracy I feel no
3:19:58Dimitri I don't want to skip over that
3:20:01I'm glad I'm glad you found meditation
3:20:02to be a powerful thing I
3:20:04it's I think it's in my mind it's one of
3:20:07the most powerful
3:20:08things I can do each
3:20:11day it's just magical it's magical it's
3:20:14magical how like or uh your room's all
3:20:17messy your brain your thoughts are all
3:20:18messy and then you just meditate for 10
3:20:2020 minutes and now it's all organized
3:20:22everything's organized and now you know
3:20:24exactly where you want to go and how to
3:20:25do it and it's just amazing it's amazing
3:20:29so don't look at accuracy profit and
3:20:31draw down we don't we're not even back
3:20:32testing yet
3:20:36though how do we even start back testing
3:20:44this we're not even back testing
3:20:48yet maybe I should start back testing
3:20:52build this
3:20:54into our back testing model cuz
3:21:00Vlad said don't even Lookout
3:21:12accuracy so I need to move this note
3:21:15here it's all coming together dog
3:21:17it's all coming together you know it too
3:21:19you're here you're present build this
3:21:22into our back testing
3:21:24models I don't know whatever cuz FL said
3:21:27don't even look at
3:21:29accuracy profit draw down obviously so
3:21:33that's profit draw down it's calar ratio
3:21:36I
3:21:36believe what is calar ratio I believe
3:21:39it's profit draw
3:21:41down draw down but yeah why wouldn't we
3:21:44be testing on actual data funds maximum
3:21:46loss
3:21:47maximum loss from Peak to tro over a
3:21:50given period not quite but
3:21:54whatever um or
3:22:05Direction and then we can use some
3:22:08things from the past to enter well
3:22:18or
3:22:19Direction then layer
3:22:22in straps do enter well ah not going to
3:22:28flush that out right now
3:22:30but I think that's the next Play From
3:22:33Here I'mma I'mma update
3:22:37this test three damn we haven't got test
3:22:40two
3:22:41yet get back testing. py into the
3:22:48equation and back test on OSS o o
3:22:56data and yeah
3:22:59yeah super dope Vlad much love to you
3:23:02what's up shorty what you
3:23:06doing
3:23:08hey hey glad my dog's been quiet this
3:23:11whole time and she just started
3:23:13congratulating you she's said woof woof
3:23:16she said woof
3:23:19woof such a good girl much to you do I'm
3:23:23saying you puppy Vibes I'm send you
3:23:25puppy Vibes dog back testing is the most
3:23:27important part and most complex I have
3:23:29great tools for it nice what type of
3:23:31tools you use I just use back testing.
3:23:33py I just use back testing. py which is
3:23:37wild
3:23:44old okay um um see what else we got
3:23:49here this is this is comparing our other
3:23:53two get big body get get them get him
3:23:57that's a big body bag right there
3:23:59somebody just got taken out on a
3:24:00stretcher BTC let's go look at it I love
3:24:05seeing these body bags no offense when
3:24:08it's on the short side I love to see it
3:24:10yo Zeus what up good to have you back
3:24:11bro it's been a while it's been a whole
3:24:14day it's been actually hasn't even been
3:24:1624 hours so I'm glad I see you every day
3:24:19now much love how's it going I'm doing
3:24:21great dog um we're just we're just
3:24:24hacking our way through these hidden
3:24:26Markoff models and got mad ideas got mad
3:24:30ideas I'm just watching somebody get
3:24:32body bagged on the chopper here they
3:24:35just got liquidated for 341 thou y'all
3:24:38probably can't even see the numbers on
3:24:40here
3:24:43huh yo is this screen stuck
3:24:51no it's not stuck is
3:24:53it can you see when I move it around oh
3:24:56my God how long has the screen been
3:24:58stuck
3:24:59for how many
3:25:03hours that's
3:25:07wild how many hours has this screen been
3:25:10suck for
3:25:19all right well I think
3:25:22that I think I got a
3:25:24bounc we're going to look I'll talk
3:25:26through this I'll talk through this I'll
3:25:27talk through
3:25:28this long time landlord says yo ain't
3:25:33nobody said
3:25:37nothing what's this a podcast
3:25:41now Jane says she's listening not
3:25:44watching
3:25:54okay so it was fine before okay maybe it
3:25:56just happened
3:25:57recently
3:26:00um I'm just going to end it with here
3:26:02and then I'm going to go grab some food
3:26:04and then I'll be back live later just
3:26:06because I got to reset this or something
3:26:09um Jane's got
3:26:12meetings she's getting ready
3:26:17certainly I'll compare this new model
3:26:19using Stoke RSI camo and adx with the
3:26:22previous models we've
3:26:25discussed yo have a great day Jane much
3:26:28love to you always
3:26:30777 landlord said about 20 minutes okay
3:26:34that's not bad Zeus I'm going to I'm
3:26:36going to restart this here soon I just
3:26:37want to compare these
3:26:44models so new model is 41 previous 97 98
3:26:4889 new model likelihood 3us 37k previous
3:26:53model minus 30 and the best one was 142
3:26:57the new model has a lower log likelihood
3:26:58indicating poor fit okay so this model
3:27:10sucks this model sucks that's okay
3:27:14that's good that's a shot that's a shot
3:27:17we took and um at the end of the day and
3:27:20I know you can't see this so it's kind
3:27:21of awkward
3:27:23but at the end of the day
3:27:27this Stoke comma
3:27:31unfortunately doesn't work
3:27:39well you
3:27:42know I'll let Vlad go ahead and say his
3:27:45seven steps though
3:27:47I'd love to hear it I'd love to hear
3:27:49your seven steps because you've been
3:27:51dropping dimes today other than that I'm
3:27:55Aude all right so I've tried seven
3:27:58different hmm which are hidden markof
3:28:02models and the best one so far is model
3:28:042 so model two was there another good
3:28:08one this Stoke did not do well so we
3:28:11passed in some different indicators oh
3:28:14that's a good idea right there so model
3:28:162 is the best what are the main
3:28:18predictors here it was Stoke
3:28:23here let's see let's see no it was um it
3:28:26was a volume change so this is the best
3:28:29thing for these inputs thus
3:28:32far best I don't know how to define that
3:28:35but it is what it is let's go ahead and
3:28:39see okay this is 0 41 but that that
3:28:41model didn't do well what is our second
3:28:44best model yeah from what you've seen
3:28:48from what
3:28:51you've seen which are our top
3:28:57two two
3:29:00modelas so model two okay this model
3:29:04consistently outperformed others with
3:29:07high State prediction accuracy 89% but
3:29:11then the lowest log or the highest log
3:29:14and the lowest Bic
3:29:17the second one is the linear linear
3:29:20regression macd and true Range model
3:29:24this model showed good performance
3:29:26though not as strong as number one so
3:29:29let's go ahead and say so far the best
3:29:33one is this model 2 where can we put
3:29:36that let's just go ahead and put a date
3:29:39here dat it I don't know the thing is
3:29:42his time is irrelevant
3:29:44so let's go ahead and say
3:29:48820b
3:29:52and Top Model is model to best obviously
3:29:58and
3:29:59second is Lyn rag Mac TR
3:30:05okay let's go ahead and just say hey
3:30:08okay well let's test this on how to
3:30:10sample data for both of them um
3:30:16update the above code in order
3:30:27to test model 2
3:30:30verse the um and model 2 and
3:30:36the
3:30:38linreg Mac one as well
3:30:43because you said it's second best this
3:30:49is oos data where are we dude how do we
3:30:53get
3:30:53here what am I even saying at this point
3:30:57oos wow wow wow wow wow wow wow wow wow
3:31:02wow wow this is why I don't like coding
3:31:04B it's a whole different language that's
3:31:06it's they speak in a different language
3:31:08y'all speak in a different language too
3:31:10it's just it's just like any industry
3:31:12though there's jargon so just got to get
3:31:14with it or get lost essentially and they
3:31:17had me lost for a long
3:31:18time okay this is O data L ma above
3:31:26please what is
3:31:29the please please send new code back
3:31:36that will test these
3:31:40models ah man I feel like I'm missing
3:31:43something there she's going to need
3:31:44something else but I'm kind of BL
3:31:54certainly I'll update the
3:31:59code model l r so these aren't quite it
3:32:03but okay I see I could have just done
3:32:05this myself really but it's okay why
3:32:08would I type why would I waste my I only
3:32:10have so many keystrokes left in my life
3:32:13I'mma save them I'mma save
3:32:16them o testing this one here
3:32:23is oh that's testing 2 it's the same
3:32:25thing let's see how much it changes 78
3:32:2979 okay what did I have up here nothing
3:32:33perfect so this isn't quite right but
3:32:37that's okay I just have to say something
3:32:39in front of it model two scaler two I
3:32:42said trained models trained models
3:32:47I was about to use some of my key extra
3:32:49keystrokes right there but I'm just
3:32:51going to copy paste I mean that's a
3:32:53smart thing to
3:32:55do o s okay slope what is going on here
3:33:01dude why do you have an
3:33:05air why would you ever have an air on me
3:33:08dog but you know I got Shooters be
3:33:17Claude Claude and then
3:33:20gbt that's it dude two messages left you
3:33:25are kidding me right you are joking
3:33:28right you are a silly silly B Billy
3:33:32right that is some wild wild stuff I
3:33:35apologize for the oversight the error is
3:33:37occurring because ta. linreg function is
3:33:40returning a series instead of a data
3:33:43frame it's all good shorty don't worry
3:33:46about
3:33:47it I'm gon let you slide this
3:33:49time obviously don't do it again
3:33:53but you decide to handle it at that time
3:33:57there we go shorty that's all you needed
3:34:00all you needed a little threat little
3:34:02threat that's all you need but Claude
3:34:04Claude is Claude is
3:34:06nice I like it so far I haven't switched
3:34:08back to GPT have you have I I mean other
3:34:11than
3:34:11like N I like it I have no no complaints
3:34:15I think it's is mobbing
3:34:18too but you know I can always jump back
3:34:20on the GPT when they need
3:34:22me
3:34:24um when I need it honestly when I'm out
3:34:27I'm actually about to jump on it so I
3:34:29better stop pumping up claw because I
3:34:32guess that's the complaint that's the
3:34:33complaint I got what do you mean you're
3:34:35cutting me off dog for an hour get out
3:34:38of here dog how you going to just cut me
3:34:41off maybe pay more that's fine
3:34:47you're just telling me that you're
3:34:48burning VC cash then you're exposing
3:34:50your hand dog that's all that's all it
3:34:53tells me if you have to cut me off at
3:34:54$20 and you won't let me you won't let
3:34:57me uh run it up then that tells me that
3:35:00these AI companies are they're upside
3:35:02down dog the metrics ain't right
3:35:05B something's going on there I'm
3:35:07sniffing it I'm sniffing it why you kind
3:35:09of why you charge and then not let me
3:35:11buy more I guess that's the question
3:35:14what's going on over there
3:35:17is it super expensive to run these
3:35:19models no yeah it could be it's cuz
3:35:22you're buying chips like a bajillion
3:35:24years out I don't know I'm not edicated
3:35:26enough to have this
3:35:29argument I'm sorry but this is the data
3:35:33so let's just go look at it let's use
3:35:35the use our last couple shots here CLA I
3:35:38love you so much you got to understand
3:35:40that it's all from love I love this you
3:35:44you got me through some hard hard times
3:35:46with
3:35:46rust the gbt couldn't
3:35:49do she couldn't do it and you could
3:35:55dude let's uh use one of our we only
3:35:59have one message left and this is going
3:36:00to be the end
3:36:02so the end of me and
3:36:06Claudia I'm going to make it a good one
3:36:08so I'm going to put a bunch of questions
3:36:10in it yeah let's run it up dude let's
3:36:11run it
3:36:12up one
3:36:18please follow the
3:36:22following follow the steps follow the
3:36:25following
3:36:26steps
3:36:30one help me
3:36:35Analyze This
3:36:38return print
3:36:41out and the chart to understand which
3:36:46did
3:36:48better in
3:36:51the out of sample
3:36:57data why did it do better I don't know
3:37:03this is a lot of pressure to have a
3:37:05limit on I can't just I can't just be
3:37:07thinking like that come on I'm closing
3:37:10this I'm gonna put a bunch of good
3:37:11questions in here though because I got
3:37:13this question ask AI
3:37:19get that up in
3:37:21there PA
3:37:23it paste that in what else what are we
3:37:27are using go
3:37:29yep yep there's another good question
3:37:31see just stack the brain
3:37:34dude I I mean I would smash the me of
3:37:37the past not just only because I know
3:37:39better python but just like we have ai
3:37:42now my guy they gave us way too much
3:37:44power
3:37:47but only coders can use
3:37:50it if this was widespread if this if
3:37:53they get if everybody could have access
3:37:55to the power that we have access to with
3:37:56python and AI or coding and AI in
3:37:59general they would shut that stuff down
3:38:01quick they just don't
3:38:05know they don't know dude we can launch
3:38:08a startup in an
3:38:11hour like a SAS
3:38:15we're literally sitting here
3:38:17in building alos all day
3:38:21long I know I'm just getting started
3:38:24know I don't know anything
3:38:26yet I'm only 4,300 hours into this game
3:38:29and I'm not saying anything till I'm at
3:38:3210,000 I'm not saying anything I don't
3:38:35have to say nothing I'm just going to
3:38:37code you can watch the code and feel the
3:38:41Vibes send love 247 777 maybe not 247
3:38:45but if you really want to watch me 24/7
3:38:46you
3:38:48could think about that
3:38:53bro that's all love it's all
3:38:56love all right use multiprocessing to
3:38:59train models okay I don't really want to
3:39:04take a hard pivot but I might need this
3:39:07I might need
3:39:11this how
3:39:13would dang I was excited to I was
3:39:17excited to get get this going
3:39:20today I'm going to use gbd4 mini yeah EP
3:39:28it see see how she can get down you
3:39:33know let's see if she can
3:39:37move all right we're going to gbt for
3:39:39mini
3:39:41and I'm not even going to ask this
3:39:43question well
3:39:46man no I am going to ask this question
3:39:49let's get one more good question then so
3:39:52what else is there use multiprocessing
3:39:54we are using gazi and hmm good ask that
3:39:57predict price
3:40:01um or send me back some code
3:40:09that
3:40:11predicts the state change and then uses
3:40:16back testing.
3:40:18py like the
3:40:22below the attached the
3:40:25attached gript does in order to combine
3:40:32processes or whatever the most efficient
3:40:38way of doing this is the
3:40:42goal is to be able to to test
3:40:48the a
3:40:50strategy where it
3:40:53switches from where where when when
3:40:58price
3:41:00switches from not price
3:41:04when the
3:41:08state switches from
3:41:12one state to
3:41:17another buy or sell based on what you
3:41:22see in the image
3:41:27attached make
3:41:30things uh use the
3:41:33optimizer use
3:41:36Optimizer like the attached code
3:41:41shows so that you can test
3:41:46for all types
3:41:49of State changes okay and
3:41:55whether to buy sell or hold I don't know
3:41:59dog I'm asking the world of her right
3:42:01now why wouldn't I why wouldn't
3:42:06I now we got to figure out where that
3:42:08code's at because I was making big
3:42:10promises over there I was making very
3:42:12very big promises and I got to figure
3:42:14out where the codes at
3:42:16do so just get up in the little hyper
3:42:21liquid action here push a little code
3:42:24here and there can cancel because that's
3:42:26not going to open Visual Studio code
3:42:29okay it's already open the now I want to
3:42:32just see what type of back test any back
3:42:35test that looks good that has Optimizer
3:42:38in it that
3:42:43runs good thing we do this a lot
3:42:47right that's a good thing we do this a
3:42:51lot
3:42:57okay there we go okay let's run it let's
3:43:00see what you got
3:43:02dog oh she can't even handle this she
3:43:06couldn't handle that paste right there
3:43:08where you at oh you're down there okay I
3:43:12was going to
3:43:13say all right so I threw a back pest in
3:43:17there asked a bunch of questions from
3:43:25her Ox Auto says is this
3:43:29live it's not it's crazy pre-recorded
3:43:33though DUS Monta says how do I find the
3:43:37moonshot coin before it explodes and a
3:43:40bot to take trade quickly any video you
3:43:42would suggest none of them bro nah no
3:43:46moon shots around here dude actually the
3:43:48sniper I guess was kind of a moonshot
3:43:50but but um back in February I was
3:43:52building the salana sniper if that's
3:43:54what you're referring to 777 OE what up
3:43:58bro good to have you here I'm asking the
3:44:00biggest I'm asking the most out of out
3:44:02of whatever her name is right now and I
3:44:04just did
3:44:06because it's my last
3:44:10chat it's the last chat
3:44:15that's the
3:44:16finale so certainly let's go through
3:44:19your questions step by step one
3:44:24analyze analysis of the out of sample
3:44:27data results the model volume volume
3:44:30change BB width volatility performed
3:44:33significantly better on the out of
3:44:35sample data with a likelihood of 3859
3:44:39compared to 27 798 for the Lin regge
3:44:44model
3:44:45a higher log
3:44:48likelihood indicates a better fit to the
3:44:52data looking at the chart the volume
3:44:55model shows more varied and frequent
3:44:57State changes which might be capturing
3:45:00more nuanced Market
3:45:03behaviors the Lin rge model seems to get
3:45:06stuck in one state state six for most of
3:45:09the later part of the data which
3:45:11suggests it's not adapting well to
3:45:13changing Market I
3:45:17conditions the volume model likely
3:45:20performed better because volume changes
3:45:22can be a strong indicator of Market
3:45:24sentiment and potential price movements
3:45:27the combination of volume volatility and
3:45:30price range BB width captures different
3:45:33aspects of the market
3:45:36Behavior it seems more responsive to
3:45:39changing markets as evidenced by the
3:45:42frequent State changes approach to ml
3:45:45for trading okay so this was the first
3:45:46part and that was
3:45:48interesting
3:45:50um you
3:45:53know I think the winner out of this
3:45:55whole test so
3:46:00far is modeled
3:46:03to that volume
3:46:06percent
3:46:10change
3:46:13is producing
3:46:16best
3:46:17results okay there we go I just got so
3:46:20much in that read me
3:46:22dude proach to ml you're on the right
3:46:24track H that feels
3:46:27good feels good you're on the right
3:46:29track the process of testing different
3:46:31indicators features and models is indeed
3:46:33a crucial part of developing machine
3:46:35learning strategies for trading there
3:46:37are some key points feature selection
3:46:39and Engineering is
3:46:41critical testing different combinations
3:46:44helps identify y what's most predictive
3:46:47for your specific
3:46:49problem when predicting price
3:46:53directly while predicting price directly
3:46:55is challenging predicting Market regimes
3:46:57or optimal strategy selection can be
3:46:59very
3:47:01valuable it's important to have a clear
3:47:03objective maximizing returns minimizing
3:47:05draw
3:47:07downs and to evaluate models based on
3:47:09these criteria not just statistical
3:47:12measures consider using other models
3:47:15like random Force gradient boosting or
3:47:18neural networks including rnn's and
3:47:21lstms for time
3:47:25series I
3:47:29mean
3:47:30yes you are on the right
3:47:35track cool just want to make sure I'm on
3:47:37the right track you know on the race I'm
3:47:40racing but I want to make sure I'm
3:47:41racing the right way look into libraries
3:47:43like s learn t flow and P torch for a
3:47:46wide range of ml models suitable for
3:47:48time series data hmm
3:47:51variations
3:47:53okay got some more Alpha
3:47:57here so hmm variations
3:48:15dang I can't see it okay here we
3:48:18go
3:48:21discreet
3:48:23hmm higher AR kle
3:48:28hmm
3:48:32factoral
3:48:34factorial
3:48:36hmm input output
3:48:41hmm
3:48:42W you might also consider consider other
3:48:45state space models like cman filters or
3:48:48particle
3:48:55filters oh goodness
3:49:00sorry oh the Inner
3:49:03Circle this your
3:49:06shot that was your chance that was your
3:49:09chance what's going on
3:49:12here so
3:49:17looks like a wild day everybody's
3:49:19getting liquidated both
3:49:23sides I want to feed in liquidation data
3:49:28dog feed in
3:49:30liquidation data and Clump it to be
3:49:36hourly
3:49:38Daily 4
3:49:40Hour Etc dude that's going to be far you
3:49:44know that's going to be far that's why
3:49:46you here dog you're like yeah that's
3:49:49going to be far that's going to be
3:49:52far it's for sure going to be far let's
3:49:54start that
3:49:58now yo 777 blessings to you dude and
3:50:02drct or Mr tomur re what's up
3:50:06dog uh can you teach me what you
3:50:10do
3:50:13um yeah
3:50:15I can for
3:50:18sure hey man he said he said he is
3:50:21literally teaching in every video watch
3:50:23all the old videos yeah dude that's the
3:50:26way that's the way just watch the old
3:50:27videos pick out what you want um good to
3:50:30have you guys here much
3:50:32love Jason hey dude how are you I'm
3:50:35doing I'm doing well thank you thank you
3:50:36very much we uh working on these hidden
3:50:41markof models right now and um
3:50:45man I love this
3:50:48stuff I love this stuff
3:50:53dude find some good ideas all right
3:51:00um we ran out of credits here
3:51:04but this is going to be interesting
3:51:06here's sample code that combines hmm
3:51:08State predicting with back
3:51:11testing all right so let's see how this
3:51:13finale is
3:51:15go
3:51:17to hmm
3:51:24models dang I knew something was Eerie
3:51:27around
3:51:29here I haven't had a music on bro
3:51:33wow spooky
3:51:35hours like what is going on dog
3:51:47e this was this
3:51:54was I'm going to listen to a little I'm
3:51:57going switch it up a little
3:51:58bit oh it's not on my computer you nerdy
3:52:04nerd but yeah I'mma switch it up a
3:52:06little bit let's get a
3:52:09little something going you
3:52:12know let's get a little vibe
3:52:15going all right now I got it there we
3:52:18go now we can hear it
3:52:22okay um feed in liquidation data and
3:52:25Clump it into hourly 4 hour so let's go
3:52:29ahead and see if we can get this back
3:52:31test uh hmm bt. py this is a new Avenue
3:52:37here we just took a left here not full
3:52:40left but here's the sample code let's
3:52:43just see what it does
3:52:45does it work we off bat I got to put the
3:52:47data in let's not hold it not hold it to
3:52:50that high of a standard no way this
3:52:51ain't going to work but maybe hey this
3:52:53is looking nice where's the day to go
3:52:56dude back test cash hmm
3:53:02strategy M
3:53:05cash why does it say it twice though
3:53:09that's
3:53:10interesting I don't know what's going to
3:53:12happen here if this is going to work but
3:53:14let's just work our way through it final
3:53:16results is
3:53:18stats they okay so yeah
3:53:25yeah let's just walk through it because
3:53:28I can't even ask oh your data path here
3:53:29we go I was like what is going on dude
3:53:33load data but I want to see
3:53:35the yeah so I don't know if this is
3:53:38going to do
3:53:40it not quite I what I need I didn't
3:53:42explain it well enough it's always me
3:53:45that's
3:53:47okay run back test run
3:53:51optimization Run final back test with
3:53:53best parameters okay let's just read
3:53:55through it see what we can get out of
3:53:56this load and pre-process data okay we
3:53:58got that we know that we load the data
3:54:02in so I'mma put the data here but the
3:54:05thing is dude is I want I want different
3:54:10data this is out of sample data here
3:54:25I don't know I got to figure out how to
3:54:27piece these
3:54:28together but we we got let's just move
3:54:31forward let's just move forward I'mma
3:54:32put this data in there for now copy
3:54:37path past
3:54:41in okay and then let's and get the first
3:54:45ER whatever that first error is let's
3:54:48get it no clothes what are you talking
3:54:50about no
3:55:02clothes I don't have ai on my side
3:55:04anymore yes I
3:55:07do above is the sample
3:55:13data of the CSV
3:55:17attached uh
3:55:19CSV for the code attached I got to say I
3:55:23like the desktop app big
3:55:26time I like the
3:55:30app I don't want a summary
3:55:34dog maybe she's going to do a summary
3:55:39and what did I put in there dude that
3:55:41was stupid of me I'm sorry I put the
3:55:43wrong thing in in
3:55:45there so I'm sorry AI GG first minute
3:55:49back I'm already dogging you no way it's
3:55:50me it's always me that's why I love this
3:55:53game so so
3:55:54much oh you lose money that's you oh you
3:55:57make money that's
3:56:00you it's crazy dude it's a good
3:56:04game the error you're encountering in
3:56:06case the code is trying to access a
3:56:08column name closes in your data frame
3:56:10but the data doesn't exist typically
3:56:12happens the CSV doesn't match what are
3:56:13you talking about we have it I
3:56:18thought is it because I have the file
3:56:21maybe because I have the file let's see
3:56:23if it drops
3:56:26it let's see one time
3:56:31no I think it's because the
3:56:37B
3:56:39CSV is a bit
3:56:41quirky and has an addition
3:56:46additional row we need to
3:56:49drop B like it's
3:57:00hot all right let's run
3:57:10it let's see what's Gucci
3:57:17wow you're really thinking about it
3:57:18that's that's
3:57:23good yeah but I got to have the code in
3:57:25there too you dumb
3:57:27dummy nah she's know she knows the
3:57:30code oh man it's not a good first show
3:57:35showing
3:57:36here but let's go ahead and uh make sure
3:57:40that we get everything that we have in
3:57:41here because remember we started with
3:57:43the uh clock and now we're over on
3:57:46GG send back full code send back fixed
3:57:53code with correct
3:57:56paths uh
3:57:58ideally ideally we train on the data
3:58:04that is above and then we save the model
3:58:10no no I have a model I already have it
3:58:12trained dude let's go look at it let's
3:58:15go get it trained model OKAY model two
3:58:19model two copy path copy path GG are you
3:58:24here are you here to stay we'll
3:58:28see I want to scratch
3:58:30this no and when we save the model
3:58:34ideally we train on the
3:58:37data which we already did and now we are
3:58:43just using using the O data below in
3:58:47order to test the
3:58:52model I guess I don't understand why we
3:58:58need to train the
3:59:02model again when I
3:59:06have one
3:59:09already
3:59:10trained I just now
3:59:14want to use it in a back test and be
3:59:19able to change the variables of the back
3:59:25test this is a the back
3:59:29test the back test should be testing
3:59:34what happens when the bot
3:59:40buys at state
3:59:44two and then cells at State four with
3:59:50changeable
3:59:52variables making those making
3:59:55those making those
3:59:59variables easy to change
4:00:11okay okay here's the other data
4:00:14um scaler to copy
4:00:20path what else do you
4:00:23need OS
4:00:28data copy
4:00:35pass sample of os
4:00:40data CSV okay let's show it
4:00:47all right let's see what you got gg40
4:00:49mini whatever your name is let's see
4:00:51what you
4:00:53got we need
4:00:55you okay to utilize your pre-trained hmn
4:00:58model and scaler of Al s day for back
4:01:01testing here's how you would modify your
4:01:05script load the trained model and scaler
4:01:09from the provided
4:01:12paths process the oh as data to use the
4:01:15same features as used in the training
4:01:22phase process the OS data to use the
4:01:25same features as used in the training Fe
4:01:28phase perform the back test use the
4:01:31loaded model to predict the states and
4:01:34evaluate the strategy
4:01:47try to prepare features predict stat
4:01:50using pre-trained model find the
4:01:52strategy for back
4:02:02testing okay she just needs a little
4:02:04help with the back testing
4:02:12part It's Tricky GG don't
4:02:21notice uh above is a back tester I use
4:02:28and
4:02:30it it
4:02:34works please use it as a template in
4:02:39order to
4:02:42recode the above
4:02:46just recode it for me
4:02:58please all right hmm models BT okay this
4:03:02is where we're at attempting to back
4:03:04test the open out of sample data I
4:03:08missed the start what's your main goal
4:03:09of this
4:03:11rewind how did you train your model or
4:03:14did you download pre-trained
4:03:19models um I don't know
4:03:22honestly I don't know the answer to
4:03:25that these are pre- Trend pre- train
4:03:28model loaded the hmm oh no I trained my
4:03:30own
4:03:31model yeah now it's
4:03:34pre-trained I think that's the answer
4:03:36I'm sorry sorry I can't answer yeah but
4:03:40thanks for getting that in my mind
4:03:46okay let's pass it
4:03:48in okay so let's see here how does this
4:03:52look now dude okay we're passing in the
4:03:55model path okay this is the path model 2
4:03:58is the best scaler 2 is the best okay
4:04:01the model equals jb. load I don't know
4:04:03what that does dude but it's okay I
4:04:05don't need to know everything this is
4:04:07the out of sample data this is out of
4:04:09sample
4:04:11data it reads it in okay okay process
4:04:14the
4:04:15data okay but what about ta dude what ta
4:04:19are we
4:04:29using that's why we
4:04:31look it's why we look at it dude it
4:04:34doesn't matter though cuz you know you
4:04:37know how that goes we got unlimited
4:04:39unlimited uses over here at open AI I
4:04:42guess we open AI guys again look at that
4:04:45we switch up
4:04:47quick I don't care dog just get the job
4:04:54done let's get it done dude okay print
4:04:57out the best optimization I I'm I'm not
4:05:00I'm not I'm not too confident in this
4:05:02but it's okay CU we'll just keep going
4:05:05so Define the strategy Define by state
4:05:08Define cell State I like this actually
4:05:11how did we get here I don't know
4:05:14last state none
4:05:16okay yes there we go buy and then sell
4:05:19and you can see it sets it up
4:05:22correctly but we don't have a stop loss
4:05:24why would we have a stop
4:05:32loss I think that I can just remove all
4:05:35of this be honest but let's just let it
4:05:37let's let it do it Thug
4:05:39fizzle let's add one more zero there
4:05:42because
4:05:50no why do I even have this dude I'm
4:05:53taking it out I'm just going to take it
4:05:55out
4:05:56sorry look at that your boy's learning
4:06:00maybe I don't want any of this
4:06:05here I'm just going to take all this
4:06:08out no stop
4:06:12loss wait wait
4:06:21I think it's just buying so
4:06:51all let's check
4:06:53it no way dude that's
4:06:56wild that's really
4:07:00wild
4:07:07989 minus
4:07:0946% cool that we got that the first time
4:07:12though
4:07:15line
4:07:1689 I mean we kind of got it you know no
4:07:20attribute called take profit online 89 I
4:07:22get it out of here I don't want it then
4:07:24I don't want it I don't want it I don't
4:07:26want it get it out of here bye byebye
4:07:29good
4:07:31night all right dude we done did it we
4:07:34done did it look at this oh how can we
4:07:38do things so fast this AI is crazy man
4:07:44it's really crazy
4:07:49dude really
4:07:52crazy so the original or the best it
4:07:55converged on was one trade pretty much
4:07:57Buy and
4:07:58Hold whoop whoop whoop fun fun
4:08:07fun and it optimized from buy state to
4:08:10four sell state to one
4:08:30are you running the code locally yeah I
4:08:34am I am I am I'm here
4:08:37locally oh what you doing boo I didn't
4:08:39even know you were down
4:08:42there I'm
4:08:45local yeah I'm running on my
4:08:48computer so this is an interesting thing
4:08:52here
4:09:00that I mean what are we testing through
4:09:03let's look at
4:09:11that what is and component
4:09:14I don't know what that
4:09:20is optimizing optimizing for n
4:09:26components yeah but I want all that
4:09:43seven that's how many states I have
4:09:46perfect perfect I was just double
4:09:48checking just double checking you
4:09:52know I reckon we all double check a
4:09:55little bit more
4:09:58but insure buy and sell states are
4:10:01different
4:10:21all right well need a better idea than
4:10:23this clearly because I don't see
4:10:25anything wrong with
4:10:27this I said buy CH from buy from
4:10:32changing one state to another and with
4:10:36the hardcoded
4:10:38stuff it's down 45 but that was like a
4:10:41total random guess
4:10:43and then the optimizer just converges to
4:10:47buy and hold
4:10:50but you
4:10:52know you can always do that that's one
4:10:56way to do this but that's not really
4:11:01trading and there's a lot of risk risk
4:11:04risis aiz says hi bro I'm a CS student
4:11:07how can I start into trading and what's
4:11:09your advice my advice is to go watch my
4:11:11oldest YouTube video and wherever you're
4:11:14at in this process find the ones that
4:11:15Vibe with you and what you learn from
4:11:18and if you ever need extra help I got
4:11:21the boot camp but the boot camp is not
4:11:23required by any means it's just
4:11:26like it's like the fast track at
4:11:28Disneyland
4:11:30bud it's like the fast
4:11:37track but it's all here it's all on
4:11:39YouTube so I want to make sure it's like
4:11:42a equal plan playing field you
4:11:45know I mean it's unequal because people
4:11:48get the fast track but it's unequal
4:11:50enough and other than that I've learned
4:11:52everything on
4:11:53YouTube like while doing it on
4:12:05YouTube all right so this is just not
4:12:07that good of
4:12:10idea but it's a good Baseline
4:12:19peran okay let's copy path
4:12:23here this is a good
4:12:30Baseline solid
4:12:32Baseline for now chopping the market up
4:12:39into regimes and then testing only in
4:12:44one regime the code
4:12:48attached
4:12:54is
4:12:58um I don't
4:13:11know is is
4:13:14predicting the
4:13:16next date and then
4:13:20just buying or selling so that can be
4:13:27useful in order to code out something
4:13:33that
4:13:38is only
4:13:40buying or selling during
4:13:44specific
4:13:50periods blow blow up says have I made an
4:13:53outgo for stocks one of my first alos
4:13:55was for stocks but I just got distracted
4:13:59by crypto and it's just more interesting
4:14:02to me essentially so it's not really a
4:14:04distraction I'm just more interested in
4:14:07crypto
4:14:08and but yeah one of my
4:14:11first or
4:14:14was
4:14:16U interactive
4:14:18stockbot interactive brokers so I got it
4:14:22in my bag let's just say that I got it
4:14:24in the bag and
4:14:27um maybe 2025 or something I'll play
4:14:30around there I don't know though to be
4:14:32honest I've been saying
4:14:34that just more interested in
4:14:39crypto and you got to do what you're
4:14:41interested in in order to do things a
4:14:43long
4:14:45time I
4:14:51amo in the boot camp do you teach
4:14:54machine learning and how to get your
4:14:55data
4:14:56sets um no I teach you how to get data
4:15:00sets though so this is a boot camp
4:15:03you're asking about and I do give you a
4:15:05data source in the boot
4:15:07camp so if you need like open high low
4:15:09close volume data for minute five minute
4:15:12whatever you need whatever you need dude
4:15:14I show you exactly how to get
4:15:17it Robo what's up dog he said my dude I
4:15:21started taking Harvard
4:15:23cs50 python class and I've been dabbling
4:15:25with my own bots in Python and also Pine
4:15:28script you're a big Ino yo I'm so happy
4:15:31you're here dude thank you thank you for
4:15:33introducing
4:15:35yourself I keep showing up if you keep
4:15:38showing up dog this is uh this is a game
4:15:41we can play Forever get always always
4:15:43find new
4:15:45edges and as our boy Jim
4:15:51said the Market's constantly
4:15:54changing the Market's constantly
4:15:57changing so there will always be new
4:16:01edges that
4:16:04appear anomalies anomalies there we go
4:16:08let's look through
4:16:10that I think this is a really good place
4:16:12we got got to because now I got this
4:16:15tool that can number one split any data
4:16:19into multiple regimes based on the
4:16:22hmm I haven't done this yet not
4:16:34done but maybe I'll do that right now
4:16:37and then to get back testing done back
4:16:40testing equation this
4:16:46now I have a great starting place for
4:16:53hmm bt.
4:16:56py as
4:16:59it allows
4:17:04for buying selling in different
4:17:08regimes just update to only allow
4:17:13trading in
4:17:19certain
4:17:21regimes okay but let's do this real
4:17:23quick
4:17:53let's see what gbt can
4:18:04do update the above hmm so that it only
4:18:12looks for four
4:18:16different um
4:18:20States I want to see if
4:18:26the results
4:18:31change send back all
4:18:40code so let's do a uh
4:18:44seven
4:18:46states and then
4:18:50uh four
4:19:01states okay these are going to be four
4:19:06states I'm going say seven regimes
4:19:18forward
4:19:19James okay
4:19:28so below is a continuation
4:19:43oh
4:19:48snap I'm using GG
4:19:51free right now so we'll see how she
4:19:56does dang claw got me hooked B I might
4:19:59have to upgrade to uh chat gbt
4:20:07again are you running multiple python
4:20:10programs Robo ass
4:20:14um yeah like you can see all the ones
4:20:17like on the screen right now what up
4:20:22programs and they collect data they'll
4:20:25do things like
4:20:28that you just got gbt plus and it's
4:20:31worth it
4:20:34nice I used to have it but then I uh
4:20:37switched to
4:20:39Cloud cuz that's what all cool kids are
4:20:41doing these days
4:20:50okay let's go ahead and look through
4:20:51this code
4:20:53here n components okay this is good this
4:20:57is good let's see if it's really good
4:20:59though this is four it should be four
4:21:02components now so the the idea here
4:21:06was we have somebody that works uh at
4:21:09the Quan firm and they say that they use
4:21:12uh four regimes risk on risk
4:21:21off I should do two regimes
4:21:26too let's just past it in there let's
4:21:29start there
4:21:30yeah let's see what's going on down here
4:21:33though state
4:21:35names update the state names
4:21:39B what is going on how many has got how
4:21:42many
4:21:43uh components does it have here is that
4:21:46what it's
4:21:47called I don't even
4:21:50know seven components dog
4:21:53what so does she just not have memory
4:21:56like
4:22:17risk on risk off what eles they say
4:22:39do
4:22:41please pretty please send back full
4:22:44code I might have to upgrade right now
4:22:47levia what's up where do I start
4:22:49learning if I want to get into data
4:22:51science YouTube dude YouTube YouTube has
4:22:55it
4:23:01all any channels Moon
4:23:05Dev Moon Dev
4:23:26this a lot of code let's see what gbt 4
4:23:28min's like
4:23:30here let's see if she can handle
4:23:39it can she do it
4:23:43oh she looking a little soft be
4:23:53honest she's looking a little
4:23:58soft I'm nervous for her that's why I'm
4:24:00looking
4:24:01through see where she done messed up we
4:24:04got risk on risk off high volatility low
4:24:07volatility all let's run it dog gbt 40
4:24:11mini she can handle it
4:24:13she can do it too
4:24:15okay that's nice that's really
4:24:19nice here we go we got four different
4:24:22colors
4:24:23now do I get access to the uh thing I
4:24:27can't believe I'm over
4:24:29here I'm over here without GPT 40 she's
4:24:33the one who started it for us and I turn
4:24:35my back I turn my back that's so whack
4:24:39that's so
4:24:41whack but gb40 minis they're trying to
4:24:43take over the world give it to me for
4:24:45free then I I'll take it for free I'll
4:24:48take it for free yeah sure sure
4:24:51sure hey but now it's missing the
4:25:02uhoh okay but it worked which is good to
4:25:06see risk on risk
4:25:08off that's all I want to know
4:25:18yo use this use this as a templet
4:25:21you you lost me
4:25:24B it looks like you lost a little
4:25:29code use the below as a template but
4:25:35just with the
4:25:38four gem Simons RI
4:25:44features features
4:25:48or States ah this is tough dude a lot of
4:25:51new stuff coming at me send back auto
4:25:54code send
4:25:56back full code nothing less
4:26:00please pretty
4:26:04please got a risk it for the biscuit
4:26:07risk on baby Robo says 777 Eric good to
4:26:11see you d
4:26:16Robo we're about to Robo we about to
4:26:19start working through gbt virtually
4:26:22I'mma ask him stuff and he's going to
4:26:23put it into his gbt and then send it to
4:26:25me oh get him get
4:26:32him I would run up your bill though not
4:26:36your bill but I be run up your credits
4:26:37mad you see I already got logged
4:26:40out oh it's about that time though 11
4:26:43minutes I can wait I can wait for it I
4:26:46might do a little meditation in between
4:26:48actually that sounds quite nice you've
4:26:51hit the free plan limit continue with
4:26:53other basic models or wow wow wow wow
4:26:56wow I didn't know they had a I did not
4:26:58know they had a upsell right there dude
4:27:00that got me that one got
4:27:04me they're goingon to they're choking me
4:27:06out here be they're choking me out I
4:27:09think I'm just going to meditate then
4:27:10because I own my
4:27:14Time come on B let's go let me upgrade
4:27:19Claud is there a is there a bigger plan
4:27:21you got a bigger plan Come on B I'll
4:27:23give you 10 more
4:27:26Claude
4:27:28pricing I'll give you a little bit more
4:27:31I'll give you 30 I'll give you 30 bro
4:27:35okay team per why you guys just always
4:27:38copy each other be creative come on
4:27:41throw this 69 on the
4:27:43price price build annually
4:27:48free learn about CLA
4:27:51team I learn about it don't know
4:27:55if gonna be about it
4:28:00but okay
4:28:03well that's too
4:28:07bad um I guess I could just get another
4:28:09membership but if I'm going to get
4:28:11another membership I'm I'm going back to
4:28:12open
4:28:14AI yeah that's what I'm feeling
4:28:18like that's how Doug said Clyde can have
4:28:21my
4:28:22babies shaking you down for that 20 Robo
4:28:28says I run mine up every day but I don't
4:28:31got experience like you Robo said well
4:28:33thank you man that's that's odd that's
4:28:35odd to think but I'm just a I'm a womb
4:28:39engineer I'm new to the game 77 7 KW
4:28:43what up bro much love to
4:28:46you much love much
4:28:52love okay so Claus got
4:28:56me Claus got me in a choke
4:29:00hole so I think I'm just going to
4:29:03meditate and come back
4:29:09later use hmm to predict four V regimes
4:29:15and then determine risk on or risk off
4:29:18okay let's just start there because I
4:29:20have these risk on risk off here and
4:29:22these are the four regimes right this is
4:29:25what we
4:29:28created let's go ahead and see what's
4:29:32Gucci with
4:29:34it so okay now it has four different
4:29:38regimes here and it says risk on riskof
4:29:41High Vol volatility low volatility but
4:29:45remember remember those don't really
4:29:47mean anything they're just four
4:29:49different things here people are getting
4:29:50licked right now
4:30:01um did we
4:30:03already test this let's see
4:30:08here execution complete
4:30:12I don't think we did the second part of
4:30:15this
4:30:17yet because we got cut
4:30:19off we got cut off train the hmm analyze
4:30:24States Save State
4:30:36changes let's go to our best one here so
4:30:38this is the B best hmm so far and the
4:30:42idea I wanted
4:30:43to tap into today
4:30:49was splitting up to four four different
4:30:54states maybe even
4:30:57less no I think four would be
4:31:02appropriate okay so here's the the best
4:31:04one here I just want to see the data
4:31:06that I get back I get this stuff back so
4:31:09hmm likelihood State prediction all that
4:31:11good stuff so that's what I need back
4:31:14here and if I go to the best here best
4:31:17is here and then I'm just going to say
4:31:21um State prediction okay I need all this
4:31:25code
4:31:27dog so I'm going to copy all this
4:31:31in to Claudia and say what's up hey
4:31:37there I have
4:31:40attached some
4:31:42[Music]
4:31:43um
4:31:45code of an hmm which is about 250
4:31:52lines of
4:31:54code use that as a template to update
4:31:59this the other the other
4:32:03script it is missing a few
4:32:09prints and
4:32:13model
4:32:16saves send back full code right back to
4:32:21it CLA Cuts us off and it slows us down
4:32:23for a second I think I'm just gonna I
4:32:25think I'm just going to buy open AI as
4:32:37well I'm just careful with that
4:32:39subscription addiction
4:32:43careful I'm
4:32:46careful but we'll see today if it cuts
4:32:48us off I might have to just light it
4:32:50back up just run it back
4:33:00up yo what up Nicholas how are you 777
4:33:04Aqua what's up bro he says yeah Aqua
4:33:07says yeah it's weird paying for
4:33:08something and still being cut off AI
4:33:10products are interesting
4:33:12bro drop your script PL says okay so
4:33:16I've dropped it it's dropped I dropped
4:33:19it on the floor I dropped it on every
4:33:21single YouTube video but let's keep it
4:33:24going B much love much
4:33:26love
4:33:28777 all right so I understand you'd like
4:33:30me to update the second script from the
4:33:33past 2.txt using the first script from
4:33:37paste. txe interesting okay so they'
4:33:39have paste and then page two as a
4:33:42template adding some prints and model
4:33:43saves I'll do that and provide the full
4:33:46updated code added more print statements
4:33:49included additional
4:33:51analysis all right let's check it out
4:33:57copy there we go that's looking
4:34:00better it's looking a lot better and
4:34:03it's got the save yes there we
4:34:09go wow
4:34:12they really gave us too much power here
4:34:13with this AI dog too much power that's
4:34:17why they had to cut us
4:34:19off yeah duh I'm just going to I'm just
4:34:21going to get open AI back and I'll just
4:34:24I'll be running
4:34:26both cuz if it's going to come me off
4:34:28every
4:34:33day I wonder if it's based on times too
4:34:36like times of
4:34:38day but this is our data back or this is
4:34:41our chart back you can see it splits
4:34:42into four different states they call it
4:34:47um I often refer to as
4:34:50regimes and risk on riskof high
4:34:53volatility low volatility are the four
4:34:56regimes but these names don't matter
4:34:58remember that remember that remember
4:34:59that okay here we go this is the stuff
4:35:02we really wanted to see here
4:35:13it's going to take a second it's gonna
4:35:16take a
4:35:17second there it is Boom okay and it save
4:35:19the model as well
4:35:22so that is our analysis here and that's
4:35:25pretty good I mean the prediction at
4:35:29least let's go ahead and triple triple
4:35:31no no no let's put it I need to put it
4:35:34somewhere better I need to put it
4:35:35somewhere better uh read me
4:35:44[Music]
4:35:49um live ops. MD I love it dog all right
4:35:55so let's go ahead and say 821 okay this
4:36:01is the this is for risk on risk off copy
4:36:04relative
4:36:08path all right so that's that now I have
4:36:12that let's just open Live UPS to the
4:36:14right and then go ahead and go back and
4:36:19get the best model here so the best
4:36:21model model two best and then I want to
4:36:24run this against that and just see what
4:36:27the diffys are if there are diffys or
4:36:30not if there's not then we'll
4:36:33just I don't know move on keep keep it
4:36:35moving you know keep moving forward
4:36:37that's it next idea you see the list
4:36:42all right this is fitting the model so
4:36:43it takes a little bit of time time time
4:36:48time so this one predicts seven
4:36:51different states so the reason I went
4:36:54down to four was
4:36:56because I wanted to try four and I've
4:37:01heard Jim Simons say or I haven't heard
4:37:04him but I've heard it it's been
4:37:06said that he thinks there's nine
4:37:10states but I've also o heard somebody
4:37:12say that he thought there's
4:37:14four so I did
4:37:16seven cuz I couldn't think of nine right
4:37:19away remember these names don't matter
4:37:22all right so I'm just testing
4:37:24both just testing both copy
4:37:28path copy relative path here and
4:37:33say VV the heck this is supposed to be
4:37:37our best
4:37:39model best model
4:37:42I don't expect good but you know that's
4:37:45our
4:37:46best
4:37:49and let's see let's look at this one the
4:37:52Bic I think this
4:37:55is I mean that looks
4:37:58better not better but I can't remember
4:38:01that's the thing is I can't remember
4:38:02what the the good one was but you can
4:38:04see the bicc here on our best model it's
4:38:07supposed to be negative so it looks like
4:38:09this one's not very good either
4:38:11unfortunately our best model is still
4:38:13the volume change okay H what does that
4:38:17tell me I don't know but that's my
4:38:19prediction let's see if I've learned
4:38:21because if I've learned that's that's
4:38:23awesome we've been here and uh you've
4:38:25been here dog
4:38:27so let's go ahead and compare these two
4:38:29models
4:38:33here Claudia what's
4:38:37up
4:38:39above I have two two hmm
4:38:43models one with four
4:38:48states and the other with seven
4:38:52states um
4:38:55please walk me through the
4:39:06results and compare
4:39:08them which is better okay and that just
4:39:11had me think of something what are the
4:39:14inputs we're putting into this one
4:39:16that's the question because here you can
4:39:20see it volume change and that's taking
4:39:22up most of the I don't know I don't know
4:39:24what you call it dog the brain power of
4:39:26the
4:39:27hmm the volume change is that weighted
4:39:30the most it's the most important it's
4:39:33the most
4:39:34important
4:39:37so what am I using over here am I using
4:39:39the volume change
4:39:42I am perfect perfect perfect are these
4:39:46the same these should be the
4:39:48same let's go and look at it
4:39:53Lynn
4:39:56okay these should be the same they
4:39:58should be the same let's see it they're
4:40:00the same perfect perfect so it's a fair
4:40:02match it's a fair match four vers seven
4:40:05this one has seven this one has four
4:40:07regimes
4:40:11okay let's go ahead and see what shorty
4:40:15said certainly I'll walk you through the
4:40:18results of both models and compare them
4:40:20based on the metrics provided let's
4:40:22analyze each aspect State prediction
4:40:25accuracy for State model 91 that was the
4:40:29new one 7even state model 89 the four
4:40:33state model has a slightly higher
4:40:35accuracy in predicting the next
4:40:38state this suggests it might be more
4:40:40state
4:40:42in it State
4:40:45transitions the log likelihood of four
4:40:48state is - 29k 7 state is 14K the seven
4:40:53State model has a much higher log
4:40:55likelihood which typically indicates a
4:40:58better fit to the
4:40:59data a positive log likelihood as in the
4:41:04seven State model is generally better
4:41:06than a negative one
4:41:08[Music]
4:41:11ban information Bic four state 58 daow
4:41:16and 7 State negative
4:41:1927 lower bic is better cross
4:41:23validation
4:41:25negative has a higher positive cross
4:41:27validation indicating better
4:41:29generalization on unseen data however
4:41:31both have high standard
4:41:38deviations suggesting some instability
4:41:41cross
4:41:45folds feature
4:41:54importance the four state volume
4:41:57change 97 but 94 on the seven State
4:42:02interesting a little bit more but did
4:42:04not make it better model fit comparison
4:42:06and conclusion the seven State model
4:42:09shows a significantly better fit to the
4:42:11data higher log likelihood lower Bic
4:42:14generalization the seven State model
4:42:16generalizes better to unseen
4:42:19data higher cross
4:42:24validation despite having more States
4:42:26the seven State model perform performs
4:42:30better according to Bic suggesting the
4:42:32additional complexity is justified by
4:42:35improved
4:42:36fit prediction stability the four state
4:42:39model is slightly more stable in
4:42:41predicting the next state but the
4:42:44difference is
4:42:51small overall the seven State model
4:42:54appears to be better choice it
4:42:56demonstrates Superior fit to the data
4:42:59better better generalization and more
4:43:01nuanced state
4:43:03representation the only metric where is
4:43:05slightly behind is the state prediction
4:43:08okay so the seven State still better
4:43:11seven State still better okay well I
4:43:16think we
4:43:20uh I'm going call this multi- regimes
4:43:25regim
4:43:27reimy multi-
4:43:29regimes okay risk on risk off I'm G say
4:43:32four here
4:43:43yo what happened to
4:43:45it four regimes okay I need to do this
4:43:49for all of them let's do it for two
4:43:53three two
4:43:58three turn this
4:44:03into just two
4:44:05regimes three regimes
4:44:12uh risk
4:44:14on risk
4:44:17off and
4:44:21frenzy send back all code send back all
4:44:27code frenzy that sounds
4:44:32fun a
4:44:34frenzy all right so three
4:44:39regimes py all I'm just going to start
4:44:41cooking up here I'm going say two
4:44:44regimes py you working for me today baby
4:44:47girl you working for me regimes. py 6
4:44:53six
4:44:54regimes six there's got to be a better
4:44:56way to do this
4:45:00but oh my God I'm just going to use all
4:45:03my credits here I'm going to use all my
4:45:05credits let's run it three regimes
4:45:16okay okay risk on risk off
4:45:20frenzy risk on risk off and
4:45:25frenzy that's good 95%
4:45:30killer
4:45:32wow log likelihood though let's see it
4:45:36let's see our good log likelihood log
4:45:38likelihood this is like a test how do
4:45:41you how do you remember
4:45:43this how do you remember if it's
4:45:45supposed to be high or low just
4:45:47repetition be so let's go ahead
4:45:52and say uh this one is that and what is
4:45:55it though three regimes copy copy copy
4:46:00relative path you thought I was going to
4:46:02say
4:46:04path all right send me
4:46:08um now do one
4:46:11with only two
4:46:19regimes excuse
4:46:24me yo send me back all the code
4:46:27too don't play me don't play me not
4:46:32today don't play me dude I think this
4:46:35one's better no it's not cross
4:46:39validation not looking good
4:46:41that should be positive if you look
4:46:44here see our best one
4:46:46here
4:46:51um up
4:46:54here hey where are you
4:46:59dude it was positive all these others
4:47:02are negative this one's not good either
4:47:05but it's
4:47:06uh SS
4:47:15copy paste
4:47:20it run
4:47:24it do
4:47:26this the
4:47:29same do the same but with five
4:47:34regimes
4:47:36then do the same with six regimes then
4:47:43do the same
4:47:45with eight
4:47:47regimes you
4:47:49name the
4:47:52states send back full code one script at
4:47:59a time then please ask me if I would
4:48:06like to move to the next
4:48:09one yo you know what I was thinking just
4:48:12right now is Sonet
4:48:153.5 they got to change the name because
4:48:183.5 n they don't got to do anything
4:48:20honestly they can do whatever they want
4:48:21to do but 3.5 like degrades it lowkey
4:48:25because you know open AI over
4:48:27here 3.5 is not you know not popping
4:48:33like that we're on four
4:48:36now dos dude all right let's get this
4:48:39plot out of here actually want to look
4:48:41at it first risk on risk
4:48:45off risk on risk
4:48:49off all right so now you see
4:48:53it log
4:48:59likelihood okay
4:49:07okay so this one here then live Ops
4:49:13and paste it right there and this was
4:49:16two regimes right copy
4:49:18P we got AI working over there right she
4:49:21working
4:49:22right she working right I hope
4:49:29so
4:49:31yes there you go shorty that's how we're
4:49:35going to do it from here on out
4:49:39okay so that was I believe that was five
4:49:43let's see let's see let's see what their
4:49:45names are 1 2 3 4
4:49:47five okay because the names don't matter
4:49:50remember that just remember that dog
4:49:52names don't matter we got to label them
4:49:53later I guess I guess that's how it goes
4:49:56I don't know to be honest I'm just out
4:49:58here in the ocean I swam out in the
4:49:59ocean there's no boat actually there is
4:50:01a boat Cloud but she left me yesterday
4:50:04that was so whack it was so so
4:50:09whack okay handle
4:50:14that Claudia Claudia
4:50:20Claudia all right so this one's looking
4:50:24bullish bearish sideways volatile and
4:50:29accumulation looks suitable to
4:50:32me
4:50:3692% let's wait for this to finish up and
4:50:39then I'm gonna have oh yeah you're going
4:50:42to keep
4:50:43going yes copi you you're going to keep
4:50:49going yes let's go ahead and put six up
4:50:52in there let's go check out the
4:50:54uniqueness of the names they came up
4:50:56with strong ball weak bull
4:50:58sideways weak bear strong bear High
4:51:01volatility okay we got this this is a
4:51:04good flow right
4:51:05here nice
4:51:08dude don't drop the ball don't drop the
4:51:11ball in this nice float let's go let's
4:51:13keep going what was that five copy
4:51:15relative path
4:51:20yes and then that was the last one we
4:51:25wanted let's do a nine
4:51:36to why not why not oh because they cut
4:51:41you off yeah that's why not
4:51:44but that's it is what it is I'm just
4:51:47about to run it up I'm G get them both
4:51:49my goal is to get cut off from both open
4:51:53Ai
4:51:56and
4:51:58Claudia in one
4:52:02day that would be dope that means we're
4:52:05working that means we're going hard and
4:52:08I'm sorry I'm sorry I actually feel
4:52:10better bad I feel bad I feel bad like if
4:52:14I feel bad if somebody has any hard
4:52:16feelings
4:52:18about that's being able code so fast now
4:52:22God 1 two 3 4 five 6 1 2 3 4 five 6 7 8
4:52:30all right let's do it let's run it don't
4:52:34drop the ball I said don't drop the ball
4:52:36keep it going did we get six on there I
4:52:39hope so
4:52:43six we
4:52:50didn't okay so let's run six
4:52:55first it's going to take a little bit
4:52:57longer I believe I don't know
4:53:09though how's this looking whoa whoa whoa
4:53:11whoa
4:53:28whoa you can print out 10
4:53:31now get copy
4:53:41nine regimes.
4:53:44py okay and then please please print out
4:53:4810 please print out 10 okay and we got
4:53:54I'm not going to show you the plot on
4:53:55this one sorry I'm just closing
4:53:57it so we're on six I
4:54:05believe so let me go ahead and grab the
4:54:08copy path here
4:54:12five and then let's just wait for
4:54:15it let's make
4:54:2110 10 regimes. py now just wait on six
4:54:26so these look like they're getting
4:54:28longer every time I guess that makes
4:54:31sense I guess that makes sense
4:54:3495 okay look at this cross validation
4:54:36score
4:54:40take shots
4:54:42dude so six six is dirty but I don't
4:54:46know I I don't trust me on that I have
4:54:48to ask a i
4:54:51because you
4:54:53know they know
4:54:56everything so this is eight and then we
4:54:58got to run eight let's run
4:55:03it and then we got 10 in the
4:55:07oven yep
4:55:11there it is I'm not going over 10 I'm
4:55:14not going over 10 that's
4:55:16just
4:55:30[Music]
4:55:33egregious there we go dang my computer
4:55:36working right now I can feel it I can
4:55:39smell it
4:55:42so this is eight we're running right now
4:55:44and then here we
4:55:47go what yo what's up
4:55:50Wishy
4:55:52anthropic
4:55:55um everything in Cosmos is
4:55:59up is Claude better than gbt
4:56:0240 uh I don't know I like it right now
4:56:05but I don't I'm indifferent
4:56:33all
4:56:38right this is taking a long time huh
4:56:42this is 8 so we'll see if nine comes
4:56:45through I'm chilling Wishy how you doing
4:56:49bro how you
4:56:54doing there we go got
4:57:02it all right let's do nine now copy
4:57:05path relative path here
4:57:11hey hey
4:57:13hey copy relative path okay perfect
4:57:17perfect perfect all right now did we get
4:57:19eight in there yeah we did we
4:57:23did nine copy path okay now I'm just
4:57:26waiting for the data just waiting for
4:57:29this fitting of the machine learning
4:57:31model the hidden markof model um let's
4:57:35see you got any questions what do you
4:57:37always mean with 777 it's just a way to
4:57:40some love and some good
4:57:43vibes Aqua says 777 symbolizes spiritual
4:57:47Perfection and divine completion
4:57:49representing the ultimate alignment of
4:57:51mind body and spirit with Cosmic
4:57:54forces wow that's
4:57:57deep that is deep I love it so this is
4:58:01the thing but I'm just going to let it
4:58:02go through perfect perfect you can see
4:58:05this one's a little less here
4:58:0788 uh how much resources does this kind
4:58:09of a I use oh that's a good question
4:58:12dude I don't know
4:58:14um I don't know I don't know about
4:58:18that everything in Cosmos is
4:58:31up
4:58:33Cosmos everything in Cosmos is up all
4:58:37right since I got a little time I'll go
4:58:40over there see what categories maybe is
4:58:43that how you do
4:58:44it oh clicked on an ad they got me
4:58:47already monetize me
4:58:56already um
4:59:07Adam yes so this is awkward
4:59:10I thought they had like a category or
4:59:16something maybe you just go to Adam or
4:59:18[Music]
4:59:22something similar coins
4:59:25Marcus how do you yeah how do you
4:59:27observe how do you observe everything's
4:59:30up in
4:59:33Cosmos that's my question so this is
4:59:36what number is this that was nine nice
4:59:38now we're just going to do 10
4:59:42copy path copy relative path here and
4:59:46then I'm going to run
4:59:51this yo Derek what's up 77 bro
4:59:55777 how are you how are you Daniel 777
5:00:00to you as well much
5:00:02love
5:00:08peace all right so this one is training
5:00:12with 10
5:00:13components 10 different components 10
5:00:16states 10
5:00:18regimes whatever you want to call
5:00:21them call them what you want to call
5:00:33them all right so here we
5:00:38go and um these are the 10 states now
5:00:54so what do you think about
5:00:56him what are your thoughts my thoughts
5:01:00are close it and let's get the data back
5:01:03and then let's get some a professional
5:01:05thoughts how about that that's a good
5:01:07idea right I think so
5:01:10we got all the data from our our file
5:01:13here compare all of these 10
5:01:20models Maybe not maybe
5:01:23not I'm going to do it one by
5:01:29one yo Wishy yeah I'm working on a
5:01:32hidden markof model um Jim Simons tipped
5:01:35me off so I just wanted to dive into it
5:01:37SP some time it's been the last few days
5:01:40still Noob as usual just a womb engineer
5:01:42over here trying to find his way in this
5:01:44deep deep ocean without a boat but I got
5:01:47a boat until the boat rugs
5:01:50me and we'll see if Claudia rugs me
5:01:54today if she does then you know maybe
5:01:57it's time
5:01:59to go beg beg for forgiveness from open
5:02:02Ai and get back on chat
5:02:05gbt because I might need both
5:02:10how's Gemini in comparison to GPD
5:02:1340 and Claude
5:02:163.5 are be curious cuz it you
5:02:23know I just be
5:02:29curious come on baby show yourself show
5:02:34yourself you're almost there so it has
5:02:3710 states so it should be a little bit
5:02:39long
5:02:43longer Daniel says thanks I've been
5:02:45discovering and testing Alo trading
5:02:47thanks to your channel my goal for 2025
5:02:50is to move from webdev to trading and
5:02:53Bot that's dope dude that's
5:02:55dope welcome to the
5:02:59world all right here we go glad you're
5:03:02here my brother let's go ahead
5:03:05and throw this 10 in there and compare
5:03:10all these just compare them all just
5:03:12compare them all except the first one
5:03:14because the first one we know is doooo
5:03:27berries above is
5:03:30[Music]
5:03:34are a
5:03:37bunch of different
5:03:40hmm
5:03:42test Iran please walk me
5:03:47[Music]
5:03:48through
5:03:51them what which did
5:03:58best
5:04:01why I mean that's that's you know it's
5:04:04good enough that's good
5:04:06enough it's good enough prompt you know
5:04:09we're deep deep deep in this game now
5:04:11and
5:04:13um you know it's been a journey this has
5:04:15been a journey I love I love these
5:04:17Journeys though I
5:04:20do I do love these Journeys
5:04:23[Music]
5:04:29bro uh oh are we back on mic
5:04:33talk is my mic messing up
5:04:36again everything clear for everybody
5:04:43[Music]
5:04:50dang I'm from the
5:04:53US West
5:04:57Coast where you where are you I think I
5:05:00think you've told me before Daniel the
5:05:03mic sounds good okay cool
5:05:10oh it's clear that's why I'm asking it's
5:05:13a um cool that's cool to hear because
5:05:17yesterday we took a pivot some uh there
5:05:20was an audio engineer
5:05:22here or somebody that sitting in a
5:05:24studio and he had some problems with the
5:05:26bass so he helped me with like
5:05:28the I don't even know honestly it was
5:05:31like learning how to code again playing
5:05:33with the little meters and stuff anyways
5:05:35it's a ATR or it's a audio tech audio
5:05:39tech
5:05:41sorry it's upside down audio tech
5:05:46Nika Cuban half Cuban half Spanish I'm
5:05:50from Spain dope
5:05:52dude great place it's a great place I
5:05:54would love to go
5:05:57back all right let's go ahead and see
5:06:01what they said so certainly I'll walk
5:06:03you through the results of the different
5:06:05hmm models and analyze their performance
5:06:07let's break this down by looking at
5:06:08various metrics
5:06:10State prediction accuracy so two regimes
5:06:12had
5:06:1497 three regimes had 95 five regines 92
5:06:19six regines okay this kind of makes
5:06:21sense I
5:06:25guess the models with fewer States tend
5:06:28to have higher prediction
5:06:33accuracy this is expected as fewer
5:06:36states are easier to predict but it
5:06:38doesn't necessarily mean they're better
5:06:40at capturing market
5:06:43dynamics log likelihood higher is better
5:06:4710 regimes was the
5:06:50best nine regimes was second best Okay
5:06:54so this just go 10 n six all the way
5:06:57down straight up this is insightful
5:07:01because it seems like it's all in
5:07:04order the models with more states have
5:07:07higher log likelihood indicating better
5:07:10fit to the
5:07:16data BC basian information
5:07:19Criterion lower is better so 10 regimes
5:07:22is the lowest so it's just the more
5:07:25regimes the better
5:07:40the Bic penalizes model complexity yet
5:07:43we see the more complex models more
5:07:46States perform better cross validation
5:07:49score higher is
5:07:53better so the more States the
5:07:56better I mean that's what it's saying I
5:08:00ain't say
5:08:01it I mean we'll see we'll see we're not
5:08:03even done reading come on moon read the
5:08:05whole
5:08:06thing again models with more States for
5:08:09perform better in the cross validation F
5:08:11feature
5:08:13importance most models heavily rely on
5:08:16volume change except for the two regime
5:08:18model which gives more weight to the BB
5:08:20W width and
5:08:23volatility based on these metrics the
5:08:25models that perform best are 10 regime
5:08:27nine regime and eight regime and we had
5:08:30we weren't even going to test those
5:08:32dude give me something with 50 regimes
5:08:38B let's see see how that
5:08:42does the 10 regime model shows the best
5:08:45performance across most metrics highest
5:08:48log likelihood lowest Bic highest cross
5:08:52validation score why did why did these
5:08:55models perform best complexity matching
5:08:57market dynamics the cryptocurrency
5:08:59market especially Bitcoin is known for
5:09:01its complex and volatile Behavior models
5:09:04with more states can capture more Nuance
5:09:06market conditions balancing fit and
5:09:10generalization while the 10 regime model
5:09:12has the highest complexity it still
5:09:15outperforms simpler models in Cross
5:09:17validation suggesting it's not
5:09:24overfitting capturing diverse Market
5:09:26states with 10 regimes the model can
5:09:29distinguish between subtle market
5:09:31conditions easy neutral bullish and
5:09:35neutral vers neutral bearish which might
5:09:37be crucial for trading strategies
5:09:40feature utilization as the number of
5:09:43regimes increases the models start to
5:09:46make slightly more balanced use of BB
5:09:49width and volatility
5:09:51features potentially capturing more
5:09:54information from the
5:09:55data however it's worth noting the
5:09:57difference in performance between 89 and
5:09:5910 seems relatively small the 8 regime
5:10:02model might be a good balance between
5:10:04performance and complexity get them out
5:10:06of
5:10:07here ldo whatever that is somebody just
5:10:11got Cho it off rest in peace don't trade
5:10:14by
5:10:15hand somebody else just got Chopper it
5:10:18off n they just got shot dang both ways
5:10:22what's going on over there that's always
5:10:25interesting when I see liquidations on
5:10:27both
5:10:30sides how did this guy get licked right
5:10:33there on
5:10:35BTC what is you doing dude right here
5:10:40what are you
5:10:45[Music]
5:10:50doing what are you doing how do you get
5:10:55licked right there how much Levy did you
5:10:57use maybe that's a mistake too I don't
5:11:00know ldo okay let's check it out let's
5:11:02see what ldo is sorry for the the switch
5:11:05up here but
5:11:09dang I don't get that maybe this data
5:11:13comes in wrong sometimes um I don't know
5:11:17but that would not be
5:11:20good somebody just got r wrecked for a
5:11:24mill on uni swap a hand Trader of
5:11:28course it's so sad to
5:11:32see may he rest in
5:11:36peace on Unis swap you were shorting
5:11:39Unis Swap and got licked let's see his
5:11:46decision okay so I
5:11:48mean shorting by hand is just a bad idea
5:11:51in general in my
5:11:54opinion all right so let's go ahead and
5:11:57say however is worth no
5:12:00noting the difference in performance
5:12:03between 89 and 10 the two regime model
5:12:05stands out for its unique feature
5:12:07importance distribution giving
5:12:10weight to BB width while it performs
5:12:13poorly on the other metrics 400,000
5:12:17Chopper out of here every time you hear
5:12:20that copter that means somebody just
5:12:21lost all that cash
5:12:24B they're on tilt they'll probably lose
5:12:27some
5:12:28more
5:12:30boy don't go back there dog don't go
5:12:34back there
5:12:41as the number of regimes increases the
5:12:44state prediction AC accuracy
5:12:47decreases this trade-off might be
5:12:49important depending on your specific
5:12:52case in conclusion the 10 regime
5:12:55performs best overall but eight or nine
5:12:58may be the give
5:13:00me please please send me the full code
5:13:06for a 15 regime
5:13:10one yo civil what's up dog good to see
5:13:14you yo Ry bear why are so many people
5:13:17hyped over Tron and their memes are they
5:13:20that good I don't know dog I don't know
5:13:22I don't know but Jane hit me up about it
5:13:25you hit me up about it right there
5:13:27anybody else hearing things about tronny
5:13:29Tron I've had bad out a bad outlook on
5:13:32Tron since you know for a long time so I
5:13:36instantly faded but then I think it
5:13:37through and it's like okay if I
5:13:39instantly fade something I got to at
5:13:41least look at
5:13:44it I'm that biased against a whole
5:13:48chain I got to at least Peak
5:13:51but I feel like they always have this
5:13:54thing every year or so they've got some
5:13:57sort of bubble of something that goes on
5:14:00um I'd be curious if other people are
5:14:02hearing about Tron and all that stuff um
5:14:05I've never really been a fan but you
5:14:08know
5:14:10that makes me want to look into
5:14:14it I get bad vibes from Tron Derek says
5:14:17I agree always have probably always will
5:14:21unfortunately so I don't know enough
5:14:26just though I don't want to dog it or
5:14:28anything but 10 regimes now we got the
5:14:3215 regime Mac
5:14:36Daddy let's see how this one does
5:14:39let's see it dance I know you're going
5:14:41to take a while it's okay I'm here I'm
5:14:45patient I'll be here for 60
5:14:47years until I can't stand no more let's
5:14:50read out these names extreme bolt let's
5:14:53see one two strong bolt three 4 5 6 7 8
5:14:579 10 11 12 13 14 15 indeed so extreme
5:15:02bear strong bear moderate bear weak bear
5:15:04bearish consolidation neutral bearish
5:15:07slightly bearish neutral slightly
5:15:09bullish neutral bullish bullish
5:15:12consolidation weak bull moderately bull
5:15:15strong bull extreme bull I can see how
5:15:16this does well now that they've
5:15:18explained it to me thank you AI
5:15:41here we go it's plotting we're done with
5:15:43half of
5:15:45it you want to see
5:15:47it I sh you do let me just copy this
5:15:51over real
5:15:52quick seven state is still
5:15:55better not
5:15:588910 is
5:16:02best all right past
5:16:05it yo civil bro you are the you you are
5:16:09the man thank you
5:16:13dude got to hop in a meeting just wanted
5:16:16to
5:16:17say hope you all a blessed day dang so
5:16:21much love to you dude 777 from me and
5:16:24everybody in this chat that is so much
5:16:26love
5:16:28dog you didn't ever have to do that so
5:16:32kind of you all right so let's see this
5:16:36this here 15 I see how this could be
5:16:45good how many days are in the year
5:16:50365 so what would this equate to 365 /
5:16:5415 24 different 24
5:16:58days 24 day
5:17:02regimes yeah 24 day regimes essentially
5:17:06let's let's not get ahead of ourselves
5:17:07looks cool colorful yeah yeah let's
5:17:10check it out though so okay okay okay so
5:17:1410 regimes what's the highest right here
5:17:16log likelihood we need High 47,000 is
5:17:19the highest oh
5:17:22snap so what if Jim was
5:17:26capping not capping
5:17:29but you know like you got to keep his
5:17:32stuff a
5:17:33secret and I don't know where I heard
5:17:36him say nine regimes but somebody else
5:17:38said four regimes what if it's really
5:17:40like 18 regimes what if it's 20 regimes
5:17:43what if it's 30
5:17:46regimes I don't
5:17:48know I'm going to keep testing this
5:17:50though yeah why wouldn't I why wouldn't
5:18:01I why wouldn't I dude
5:18:12so much love to you dog
5:18:15civil he said be for real you put you
5:18:18putting on Game for free
5:18:2099 every day got to respect
5:18:25that you're a good dude
5:18:28man you're a good
5:18:35dude you got me over here shooting
5:18:37jumpers
5:18:38he just threw five at your boy for no
5:18:41reason I'm over here shooting jump air
5:18:44jumpers probably a couple air balls but
5:18:47you
5:18:48know appreciate you man that's love
5:18:51that's love I just know the pro the
5:18:54process of pulling out a credit card to
5:18:57to send a super like that is just like
5:18:59you're spending your time which is so
5:19:03kind so kind so kind good to see you my
5:19:07dude good to see you
5:19:10so this takes longer every time which is
5:19:14it makes sense right so it makes sense
5:19:18explain why explain why it takes longer
5:19:23to run
5:19:24the um
5:19:28man this
5:19:30stuff this part of run the code with
5:19:35more
5:19:37um more
5:19:45States I just want
5:19:49to get an ai's perspective on this
5:19:52because you know me I know nothing
5:19:56B I'm just here
5:20:05vibing all right so here we go this is
5:20:07good
5:20:10the increase in runtime for models with
5:20:12more States is due to several factors
5:20:14related to the computation computational
5:20:16complexity of hidden Markoff models hmm
5:20:19let me explain the main reasons
5:20:21parameter estimation in an hmm the end
5:20:24States we need to estimate n * n
5:20:27transition
5:20:28probabilities the transition Matrix and
5:20:31initial State probabilities
5:20:33probabilities and sets of emissions
5:20:37probabilities in this case means and
5:20:40covariances for each state as n
5:20:42increases the number of parameters to
5:20:44estimate grows rapidly requiring more
5:20:47computations expectation maximization em
5:20:50algorithm the bomb Welch algorithm a
5:20:53type of em is typically used to train
5:20:56hmm each iteration of this algorithm
5:20:59involves forward backward procedure
5:21:02complexity where T is the length of the
5:21:06sequence reestimation
5:21:08maybe I asked to I asked the wrong
5:21:11question reestimation of parameters also
5:21:14grows within more States mean more
5:21:17iterations are often needed for
5:21:20convergence State sequence prediction
5:21:22the vctor B algorithm has a Time
5:21:26complexity of N2 S as n increases the
5:21:31time to predict States grows
5:21:34quadratically Matrix operations
5:21:37operations like Matrix multiplication
5:21:39have a complexity of O to the N the 3
5:21:44for n * n matrices larger State spaces
5:21:48okay so convergence time the likelihood
5:21:51landscape becomes more complex with more
5:21:55parameters this can lead to more
5:21:57iterations of the EM memory usage more
5:22:00states require more memory to store
5:22:03parameters and intermediate
5:22:06calculations if memory usage exams
5:22:08exceeds available Ram it can lead to dis
5:22:12swapping dramatically slowing down
5:22:14computations cross validation and model
5:22:19evaluation okay so it just takes longer
5:22:22sweet I'm going to get one send me one
5:22:25send me
5:22:27one one with uh I want two two per month
5:22:3324 States
5:22:40two states per
5:22:42month cuz what if that's the case you
5:22:56know so this is the new one and it looks
5:23:01like this was 15
5:23:07regime so
5:23:11more I don't know how to make my notes
5:23:14look good
5:23:22but 15
5:23:26State
5:23:28cuz that's what she
5:23:31said no it's not what she said but I'm
5:23:34just going to have it there so Bic how's
5:23:38that looking so it's obviously not as
5:23:42good
5:23:4486 83 so it just goes straight
5:23:47down that makes
5:23:49sense I don't know maybe I'm going
5:23:51overboard now but we'll
5:23:54see you
5:23:55know uh hmm it means hidden markof
5:24:00model ryar says I've always stayed away
5:24:03from Tron he did a fork of pump fun on
5:24:07Tron the volume looks interesting but
5:24:09Tron
5:24:11Tron yo what's the link it's called Sun
5:24:14pump or
5:24:15something that's just so sketch this is
5:24:23it okay um I'll look into this a little
5:24:26bit later though let me uh get this last
5:24:29one done here and then I think this is a
5:24:34pretty
5:24:35good tippity tip tip dude
5:24:39I think we learned something big here
5:24:42dude we learned something big
5:24:46today already
5:24:52dude all right so this is the 24 regimes
5:24:55let's see if it's truth I think it is
5:24:58let's run it
5:25:00dude all right so this is 24 regimes if
5:25:03you got questions let me know
5:25:18Sun pump are there any other resources
5:25:21for sun
5:25:23pump while I'm waiting on this
5:25:26Hmm this is so funny we're using machine
5:25:29learning models to predict Market
5:25:31regimes but also looking at some
5:25:36punp oh my God
5:25:39okay
5:25:40so I don't see much movement like the
5:25:42other one but what's Sun swap is that
5:25:46there that
5:25:48thing
5:26:00um one minute
5:26:16the thing is I don't know how to use
5:26:17pump fun I never I never used it and I
5:26:20can't build a bot for it I'm not that
5:26:22interested is this just because it just
5:26:26launched like if there's a ton of volume
5:26:29I don't know what is this where are you
5:26:30guys getting this volume from like uh
5:26:33not volume where you getting this uh
5:26:35this information from is like Twitter
5:26:37everybody's talking about on Twitter
5:26:38are there
5:26:41anybody yeah like I'm curious I'm
5:26:44curious anyways here's 24 this this is
5:26:47for Kobe so it should be the best yeah
5:26:50it is the best log likelihood so
5:26:53far I'm going to go ahead and say hey
5:26:58please tell me one or two things about
5:27:01my
5:27:03data please go ahead here and tell me
5:27:06this here's the 15
5:27:09here's 15
5:27:12regimes attached is 15
5:27:16regimes and 24 regimes compare them to
5:27:22my
5:27:2478 my
5:27:308910 which is best okay
5:27:34so that's going to be done here in a few
5:27:39everybody's talking about this over here
5:27:42it seems super sketch already I mean I
5:27:45think pum fun seems sketch but so if you
5:27:49get to put it on Tron then it's like
5:27:52okay A little Sketcher
5:27:54but so sketch on sketch equals super
5:27:58sketch in my opinion
5:28:09a few seconds ago here we
5:28:22go I'm a pass for now I'm G pass on any
5:28:26uh any of this I'mma fade it I'mma fade
5:28:29it I think I'm G
5:28:33fade is this a bad idea I don't think so
5:28:36I think it's I think it's a good idea
5:28:39anything Tron I instantly fade I'm sorry
5:28:44it's
5:28:47sketch I don't know why but I got no
5:28:50proof I've got no proof but see they got
5:28:55API he's over here
5:28:57like they got API
5:29:01though yeah I'm
5:29:03good I'll keep an eye on it thank you
5:29:05for
5:29:06like like it for me but you know pump
5:29:12fun seems cool like I just never got
5:29:15into it I can't use right API whatever I
5:29:20build snipers I build snipers so I just
5:29:23kind of faded pump
5:29:26fun
5:29:28um probably for the better it seems like
5:29:31I'll keep an eye on it so Tron double
5:29:36fade I'mma stick get I'mma stay right
5:29:39here I'mma stay right here I'mma stay
5:29:41right
5:29:42here let's check out the see what people
5:29:45are saying though
5:29:48maybe maybe it
5:29:52will convince
5:29:59me while we're waiting
5:30:09ton dropped an
5:30:20R all right nothing yet nothing
5:30:31yet dang this is taking a long
5:30:35time but that's good let's see what my
5:30:38computer activity is looking
5:30:40like I haven't looked at that in a long
5:30:43time that's a luxury right there
5:30:48dude remember back in the day when I
5:30:50could barely keep the stream up we've
5:30:51come so
5:30:55far I could barely keep this stream
5:30:59going now we got every order coming in
5:31:03every order on binance flying through
5:31:05our screen all the time
5:31:09add a apple good job good job much love
5:31:15so much
5:31:17love so much
5:31:25love here we
5:31:29go this might be the best one
5:31:34so I mean what does that tell me I don't
5:31:37know
5:31:4224
5:31:48State let's just let's just go ahead and
5:31:50see what she
5:31:52says oh no no no no
5:31:55no did
5:31:59that oh my
5:32:02God copy pass no no no no paste it in
5:32:05paste it in 24 States copy
5:32:08P why do I just switch up the whole
5:32:11thing from here I don't get
5:32:15that okay okay now we're good we're good
5:32:19now that's the
5:32:21path this is the path every day
5:32:28dog every
5:32:31day every day dude every day just keep
5:32:36going
5:32:44just keep going Peace by piece makes a
5:32:46little more sense every day every day
5:32:49dude copy
5:32:53path flag
5:32:55it that's a flag right that's what they
5:32:58call
5:32:59it I'm growing up in front of your eyes
5:33:02it's
5:33:03amazing thank you for all your help
5:33:07could not have done it without
5:33:11you you teach me things I don't even
5:33:13know I need to
5:33:15know attached is 15 regimes and 24
5:33:18regimes compare them to the 89 10 and we
5:33:21just keep on
5:33:25going come on
5:33:27dog I'm still here where you
5:33:32at the screen's gone is what you're
5:33:35telling
5:33:35me I was just talking about this I was
5:33:39just giving you props Apple the screen's
5:33:42gone again so that means I'm gone
5:33:50again that's crazy dude essentially it
5:33:54takes the stream
5:33:56down all right this makes sense though
5:33:58I'm might have to cut some of these
5:33:59orders off I was just giving daps to
5:34:05Apple thank you for providing the
5:34:08results for the 15 regime and the 24
5:34:10regime models let's compare these with
5:34:13your previous 89 and 10 regime models to
5:34:16determine which performs best we analyze
5:34:19each
5:34:21metric State prediction accuracy eight
5:34:24regimes not
5:34:27provided nine regimes 88 10 regimes
5:34:3286 81 okay the accuracy decreases as the
5:34:35number of regimes increases which is
5:34:37expected the more state to make
5:34:38predictions more challenging log
5:34:42likelihood higher is
5:34:48better the log likelihood consistently
5:34:50increases
5:35:0815,000 versus 68,000 log likelihood in
5:35:11consistently increases with more regimes
5:35:13indicating better fit to the
5:35:15data this keeps decreasing lower is
5:35:18better the Bic continues to decrease
5:35:20with more regimes suggesting that the
5:35:23increased complexity is Justified to
5:35:25improve to the improved fit cross
5:35:28validation score higher is better dude
5:35:33so this is just telling me that more
5:35:36regimes more better more regimes more
5:35:38better let's do 56
5:35:40regimes no no no no come on now you're
5:35:43going crazy dog 24 is
5:35:48fine what you think I should do 56 throw
5:35:50a 56 and if I should do 56 it's going to
5:35:52take a
5:35:53minute I might have to shut some of this
5:35:56stuff down shut some of these order
5:35:58flows down should I do 56 that's how
5:36:01many years are months are in the year
5:36:03right weeks in a year 52 56 it's 52
5:36:06isn't it it's 52 isn't it how many weeks
5:36:09in a year how many
5:36:11weeks in a year time is irrelevant to
5:36:15me I don't live by time I time I live I
5:36:19time my
5:36:22living 4 hours a day
5:36:26dog that's
5:36:30it I locked in I didn't know how to code
5:36:32I locked in for 4 hours a day and it was
5:36:34so hard it was so hard dog it was so
5:36:38hard boohoo me oh I have to tap my
5:36:40fingers for 4 hours a day on this little
5:36:43code and now we got AI this is so so
5:36:46hard 4 hours a day you kidding me
5:36:49dog I used to game for eight hours a day
5:36:52dude I used to game for eight hours a
5:36:55day four hours is
5:37:02easy feature importance
5:37:07interesting I like this on the 24
5:37:09regimes it goes down a little bit it it
5:37:12passes it over to
5:37:14volatility analysis model fit the 24
5:37:17regimes chose to fit best fit the data
5:37:20with the highest log likelihood and
5:37:22lowest
5:37:24bicc the 24 regime this is for
5:37:27Kobe this is for Kobe we got to use the
5:37:3024
5:37:40yeah I'm I'm sticking with it though
5:37:43maybe the maybe it's better maybe
5:37:45there's others that are better but I got
5:37:47to I got to kick it with the 24 I mean
5:37:5050 is pretty pretty N I got to do
5:37:5350 give me the code for a 50
5:38:04state that's probably GNA be too many
5:38:07but
5:38:12but let's just get it anyways let's just
5:38:14get it anyways might take a while that's
5:38:17okay I'll just run it on a terminal dog
5:38:21that we can keep it coding duh come on
5:38:25moon you got to be thinking sharp all
5:38:27the
5:38:29time Come on B tap in
5:38:38based on the metrics the 24 model
5:38:40appears to be the best performing model
5:38:43overall it shows the best fit to the
5:38:46data the best
5:38:48generalization and the the most balanced
5:38:55features this is fun dog this is fun
5:39:10all right I'm going to do 150 and we're
5:39:11just going to let this run for a while
5:39:12over somewhere else I'm not going to
5:39:14worry about it but it's free it's free
5:39:16dog this is free a free it's a free beat
5:39:20dog if it's working 24 or 50 those are
5:39:23the two most important numbers to me
5:39:32so I would love that to be part
5:39:35of part of
5:39:38a
5:39:40system
5:39:42because then I get to smile every time I
5:39:45think about it
5:39:48cond activate T flow bro cond activate T
5:39:53flow bro python run it what cussing what
5:39:58is going on copy P python python python
5:40:04python python python
5:40:09python
5:40:11whoa dude what's going on
5:40:17here con activate T
5:40:20FL okay I'm
5:40:23there copy
5:40:26path LS CD oh maybe I got to do
5:40:33that no I don't have to CD in anything
5:40:49[Music]
5:40:51python what's going on
5:40:58dog what is this right
5:41:01here is that a thing it's always been
5:41:05there
5:41:07look at it I've never seen
5:41:10that yeah I guess
5:41:15so and does it run from
5:41:18here
5:41:26oh okay I'm not quite
5:41:34sure all right big guy
5:41:37love you
5:41:42bro I don't know if the 50 is going to
5:41:44work out for us but let's try it let's
5:41:46give let's give it a try give it a
5:41:52try let's give it a
5:41:55try 24 seems acceptable though that's
5:41:58what um
5:42:02every 10 days or something 4 days 14
5:42:06days 17
5:42:08n 365 /
5:42:1224
5:42:13365 / by 24 so 15 days every 15
5:42:23days main execution this is what we're
5:42:25changing dude this is what we're
5:42:28changing okay let's check it out
5:42:32then main
5:42:34execution no we want all of that out of
5:42:37here peace all right let's run
5:42:40it all right so here we go we got 50 for
5:42:43the big guy this is going to take a long
5:42:45time let's see if I can run it from my
5:42:48other thing here copy path okay um
5:43:00cond all right we're not doing
5:43:0450 we're not doing 50 I'm deleting in it
5:43:08much
5:43:10love 24 is the
5:43:13best so if there's 24 regimes for Kobe
5:43:19that means
5:43:28that okay I got this now if there's 24
5:43:31for Kobe
5:43:37I was
5:43:40pre-shock that was pre-shock for sure
5:43:43post
5:43:46Shack
5:43:47trained I want to get one of these back
5:43:49tests
5:43:51going BT
5:43:53okay the first
5:43:59paste let started new new little
5:44:04chesy the first paste of code above
5:44:11is a template template for you to follow
5:44:17for the hmm plus back
5:44:21test I want to test the trading idea of
5:44:37I mean what idea do we want to do is the
5:44:39question
5:44:40dude hmm
5:44:47models 24 regimes
5:45:02dog this is what my I want to make a
5:45:04script here to explore this data
5:45:07no
5:45:14dude I want to make a script to explore
5:45:19the above data with Matt plot
5:45:24live um I'd like
5:45:28to print out
5:45:31pie of it
5:45:57all right so I'm plotting it out now M
5:46:00plot
5:46:05Liv all right so so from all this I
5:46:08found the conclusion is 24 regimes
5:46:12volume change
5:46:14heavy there you go there's the alpha
5:46:16dude much
5:46:19love I don't know if it's actually Alpha
5:46:21but we'll see
5:46:56strong ball okay this is sketch AF
5:47:03dog I don't like it already I'm sorry
5:47:08so this is I'm glad we're looking at
5:47:09this because this
5:47:11is this shows me okay this is looking at
5:47:15the past
5:47:19right let's ask AI let me try to get
5:47:22this thought out
5:47:29here a sample of the state changes for
5:47:35the 24 regime is
5:47:39above My worry is that each row has a
5:47:46different
5:47:49regime but
5:47:51these are all
5:47:53decided after they
5:47:58happen so how do we know
5:48:03that the accuracy
5:48:08is 80% 80% or whatever is shown below I
5:48:15do that why am I like
5:48:25this it's not whatever below it's
5:48:27literally 83% dog
5:48:39dang that's funny I didn't even do it I
5:48:41didn't even get
5:48:45it I think it's up here though
5:49:13post Haw classification you're correct
5:49:15that these regimes or states are
5:49:16typically classified after the fact this
5:49:19is common in financial market analysis
5:49:21where patterns and Trends are often
5:49:23identified
5:49:24retrospectively accuracy
5:49:27measurement when someone claims an
5:49:29accuracy at 80% or any percentage for
5:49:32such a classification system they're
5:49:34referring to how well the system
5:49:37performs in predicting future Market
5:49:40Behavior based on these identified
5:49:43States here's how to think this might
5:49:48work historical analysis the system is
5:49:51first developed in tested on historical
5:49:56data where states are identified post
5:50:02HW forward testing the system is then
5:50:05applied to new unseen data either
5:50:07through out of sample testing or actual
5:50:10real time
5:50:14application to see how well it predicts
5:50:16the future Market
5:50:18Behavior accuracy
5:50:20calculation the accuracy percentage
5:50:23would typically be calculated based on
5:50:25how often the market behaves as expected
5:50:28given a particular State
5:50:34classification verification process to
5:50:36truly verify the claimed 80% you would
5:50:39need a clear definition of what
5:50:41constitutes correct prediction for each
5:50:44state a set of test data that wasn't
5:50:47used in developing
5:50:51the classification
5:50:58system that's what we need to do thank
5:51:02you I'm
5:51:05AI Alis Le okay I want to I want to go
5:51:09back
5:51:14here send me back the code
5:51:21that yo what up 777
5:51:25Bailey uh that um
5:51:48okay so model regimes hmm models trained
5:51:51models 24 copy
5:51:58path where's the other 24 here somewhere
5:52:08copy
5:52:11P okay and
5:52:14then what was the best one before it was
5:52:18seven
5:52:20seven
5:52:21s
5:52:25seven 6 10 3 where is
5:52:30seven we on a seven here all right we're
5:52:32going to do eight
5:52:34then we're going to do eight cuz eight
5:52:36was good
5:52:38too copy path oh sorry seven is two duh
5:52:43seven is
5:52:45two model two copy
5:52:52pass
5:52:58duh copy
5:53:01path okay
5:53:04so yeah I want to do that now
5:53:07so
5:53:09now let's go up
5:53:17here now back to or not
5:53:20now doing an oos test on model
5:53:272 7
5:53:30States verse Model 24
5:53:36States
5:53:38stats okay so that's going to be good
5:53:42dude it's going to be really good I like
5:53:45the idea how about
5:53:46you how about you dude do you like it I
5:53:50like it a lot o
5:53:54s o o s
5:53:58okay o s two No 7 verse 7 v24
5:54:06py hi my name is
5:54:09Mev let's run it
5:54:13cousin all right there we go what's that
5:54:17looking like I don't know you don't you
5:54:19know I don't know I need shots I need to
5:54:22get shots up I need to get shots
5:54:24up till I can say anything I don't know
5:54:27anything in confidence I'm sorry
5:54:45um thank
5:54:47you but I also want to get my
5:54:51uh Bic 15 24 here's my 24 that I need
5:54:57because I didn't write it down and it's
5:55:00the
5:55:04best it's the best in the
5:55:07ass all right 24 past it in there
5:55:13yes let's do a little two-hand action
5:55:19here
5:55:21Bic cross
5:55:24Valley feature
5:55:26importance ball change BB
5:55:29with uhoh I'm lost to be
5:55:33there okay there we go
5:55:46now
5:55:56what this I'm going to actually run this
5:55:59again because I'm going get on over here
5:56:02and say hey I actually no this ain't
5:56:07it
5:56:11hmm did we get this script though this
5:56:15ain't
5:56:17it this ain't
5:56:20it oh
5:56:23no I lost
5:56:26it okay that's
5:56:28fine fine we're just going to run it
5:56:30from here
5:56:31then OSS o o OSS s s
5:56:43oos let's get it okay okay
5:56:47[Music]
5:56:51okay to throw extra couple KS on there
5:56:55because I was
5:56:58sketch boy sketch seven State versus 24
5:57:04State please
5:57:07break
5:57:09down the uh
5:57:11OS
5:57:13test and
5:57:16the results which is
5:57:25better so I did a OS test here copy copy
5:57:29copy relative path
5:57:33okay live Ops here
5:57:41let's go ahead and check it
5:57:44out certainly let's break it down out of
5:57:47sample uh test for the seventh stay in
5:57:50the 24 Hidden stay applied to BTC price
5:57:54log likelihood 7 State 3890
5:57:5824933 the log likelihood is a measure of
5:58:01how well the model fits the data a
5:58:03higher log likelihood indicates a better
5:58:05fit in this case 24 model has
5:58:07significantly
5:58:08higher suggesting it fits the out of
5:58:11sample data better than the seven State
5:58:13model visual interpretation the seven
5:58:16State model shows distinct broad State
5:58:20changes over time the 24 State model
5:58:23exhibits more frequent and granular
5:58:26State
5:58:28transitions the 24 State model appears
5:58:30to capture more Nuance market conditions
5:58:33potentially identifying s sulo
5:58:37subtler shifts in the Bitcoin price
5:58:41Dynamics State utilization these seven
5:58:44State model seems to use all of its
5:58:47states
5:58:49regularly the 24 State model also
5:58:52appears to utilize most if not all the
5:58:55states indicating that the additional
5:58:58complexity is being put to use in
5:59:01describing the
5:59:03data responsiveness to price movements
5:59:06the 24 State model shows more frequent
5:59:09State changes which could indicate
5:59:12higher responsiveness to short-term
5:59:15price movements and market
5:59:17conditions the seven State model
5:59:19captures broader longer term Trends in
5:59:23the market complexity versus
5:59:27interpretability interpretability
5:59:29interpreting being able to interpret it
5:59:31the seven State model is simpler and
5:59:33potentially easier to interpret
5:59:36with each state possibly representing a
5:59:38more distinct Market condition the 24
5:59:42State model offers more granularity but
5:59:45may be more challenging to interpret the
5:59:48differences between some states could be
5:59:51subtle which is better based on the
5:59:53information the 24 State model appears
5:59:55to perform better in terms of fitting
5:59:58out of sample data as evidenced by a
6:00:01higher log likelihood score it captures
6:00:04more nuanced Market Behavior and
6:00:06potentially offers a more detailed
6:00:10representation however better depends on
6:00:12your specific
6:00:14goals if you're aiming for the most
6:00:16accurate representation of Market States
6:00:19and don't mind dealing with increased
6:00:21complexity the 24 State model is
6:00:28superior if you prefer a simpler model
6:00:31that's easier to interpret and still
6:00:33captures
6:00:37major market trends the state seven
6:00:41State model might be more appropriate
6:00:44for trading purposes the 24 State model
6:00:47might provide more frequent signals
6:00:50which could be beneficial for shortterm
6:00:52trading strategies the seven State model
6:00:56might be more suitable to for longer
6:01:00term Trend
6:01:02following the 24 State models better
6:01:06fit suggests it might be more predictive
6:01:10but this should be verified with
6:01:11additional metrics like prediction
6:01:13accuracy on future
6:01:18data what other
6:01:21metrics could we print out to see which
6:01:27is
6:01:29better Jane what up
6:01:31777 good to see you good to see you Jane
6:01:37says check your
6:01:39DMs I check them I'll check them I'll
6:01:42check them I appreciate you so so
6:01:44[Music]
6:01:45much lots of licks Happening
6:01:49Here excellent question to get more
6:01:51comprehensive comparison between the 7th
6:01:53State and 24th State models we can add
6:01:56several additional metrics here are some
6:01:58important metrics we could calculate and
6:02:01print out the basian the Bic the basian
6:02:04information Criterion this metric
6:02:06balances model fit with model complexity
6:02:09lower bic is
6:02:11better
6:02:13aaii information Criterion on the I AIC
6:02:16similar to Bic but penalizes complexity
6:02:20less lower AIC is better State
6:02:24prediction accuracy how well the model
6:02:26predicts the next State Higher accuracy
6:02:30is better transition Matrix analysis
6:02:33entropy of the transition Matrix to to
6:02:35measure State
6:02:37stability feature importance to see
6:02:40which features are most influential in
6:02:44each model perplexity a measure of how
6:02:47well the model predicts a
6:02:52sample send back full
6:02:56C I'm getting the most out of you you're
6:02:58typing every letter I'm sorry I'm sorry
6:03:02I'm not sorry I'm going to run you up
6:03:03every single day until you cut me off
6:03:05and then I'm going to go get somebody
6:03:06else I'm sorry every single day
6:03:11dude CLA you know we're hitting we're
6:03:14hitting the limits every single day I'm
6:03:16not leaving till we hit the limits
6:03:18that's my new thing I'm not leaving till
6:03:19I hit the limits on both no let's start
6:03:21with Cloud let's
6:03:22start: no no but I get I be get the
6:03:25limits too early no way keep going keep
6:03:28going every single
6:03:34second for
6:04:06iby
6:04:09contents 10 messages look I'm already
6:04:12there that's funny bunny some funny
6:04:14bunny stuff think I'm just going to sign
6:04:17right back
6:04:19up I think I'm just going to sign right
6:04:21back
6:04:28up so this is going to give us a little
6:04:30bit more detail
6:04:34supposedly let's check it out though if
6:04:36that's truth or
6:04:39not oh
6:04:41man here we go we got some things though
6:04:45there we go that's what I like to see
6:04:51dude all code back if they're like
6:04:53limitting on messages number of messages
6:04:56then that's different
6:05:07have you experimented with any high
6:05:09frequency models for voting
6:05:11agents I'm new your streams introduced
6:05:13from Twitter don't know which AI to use
6:05:17if any to execute logic or if it's all
6:05:22Quant what are HF
6:05:25models not sure what that is like high
6:05:30frequency models as voting
6:05:33agents nah man you're ahead of me you're
6:05:36ahead of me
6:05:42bro let's go ahead and pop this in here
6:05:46thanks for stopping by though appreciate
6:05:48you much love let's go ahead and say run
6:05:53it up there we
6:05:55go okay this is more
6:05:58info here's more info
6:06:05oh and
6:06:07there's okay okay cool cool
6:06:12Co above is the prince out
6:06:17please walk me through it and how if and
6:06:23if it changes your
6:06:27outlook or things I should know about it
6:06:32I don't know dude I don't know dude
6:06:36I just be asking anything this is why I
6:06:38run out of chat so fast they want me to
6:06:41be more thoughtful and I'm not going to
6:06:43be I'm
6:06:44sorry certainly let's walk through the
6:06:47results for both models and analyze what
6:06:50they tell us about their performance on
6:06:52the out of sample data seven State Model
6:06:54results log likelihood of 3890 this is
6:06:58lower than the 24 State indicating a
6:07:01less precise fit viic minus 7,000
6:07:062 and the AIC isus
6:07:107, 601 this is on the 7th State lower
6:07:14values indicate better models these are
6:07:16higher less negative than the 24 State
6:07:19suggesting the 24 State model provides a
6:07:22better balance of fit let's go see it I
6:07:25mean okay so here I can see a little
6:07:28bigger
6:07:30here just so we can kind of understand
6:07:4424 State you can see the accuracy is
6:07:46much lower
6:07:47though transition entropy what the heck
6:07:50does that mean dude what does that mean
6:07:53I don't know it's okay though cuz it's
6:07:57just a
6:07:58word feure importance perplexity I'm
6:08:02perplexed you would say perplexity on
6:08:05the print
6:08:06out average State
6:08:09duration
6:08:123.59
6:08:166.3 average State
6:08:23[Music]
6:08:27duration okay we looked at them well
6:08:30let's compare them now so State
6:08:33prediction is 84%
6:08:35on the 7 verse 72 transition entropy
6:08:393.96 verse
6:08:422.41 and then that stuff doesn't matter
6:08:45I mean it does matter but State
6:08:47prediction is
6:08:4884 on the 24 indicating a better short
6:08:51term is that cap no no this is higher
6:08:55than the 24 all right transition entropy
6:08:592.41 lower than 24 State model
6:09:02suggesting stable State transitions
6:09:05feature importance the volume change of
6:09:0994% perplexity is
6:09:130.0000 very low perplexity and indicates
6:09:15high confidence in
6:09:17predictions
6:09:24nice average State duration is 6.3
6:09:27States persist no longer for longer
6:09:29periods compared to the 24 State model
6:09:36State usage 0 two 4 and six
6:09:41are most frequently used while state
6:09:44three is never
6:09:56used all right this is actually useful
6:09:59this is awful right
6:10:00here I mean maybe we'll see let's go
6:10:03check it out let's go check it out see
6:10:05what can do with this thing
6:10:09dude all right so State usage for the
6:10:1624 I want to see that for seven too why
6:10:19because then I can I can do some thg
6:10:21dizzles over here and try some
6:10:28things all right for seven cool cool
6:10:32cool um now let's says model fit the 24
6:10:36State model provides a better fit to the
6:10:38data higher log
6:10:42okay the seventh state has a a higher
6:10:45short-term prediction accuracy stability
6:10:48the seventh state shows more stable
6:10:51State
6:10:55transitions feature utilization the 24
6:10:59State model makes more balanced use of
6:11:01all features unused States
6:11:06both models have unused States 1 and 7
6:11:09and 2 and 24 suggesting some redundancy
6:11:12in the
6:11:13space State distribution the 24th state
6:11:16has a more even State Chopper him out of
6:11:19here get him out of here 400k off damn
6:11:24dude that hurts that hurts my soul that
6:11:27hurts my soul that hurts my soul rest in
6:11:30peace young
6:11:32cat changes in Outlook the the 24 State
6:11:37model shows stronger statistical
6:11:40performance better fit lower Bic AIC
6:11:42than initially thought despite its
6:11:44complexity the seven State model's
6:11:46higher prediction accuracy and stability
6:11:48could make it more practical for certain
6:11:51applications especially short-term
6:11:53predictions things to consider the
6:11:55trade-off between model complexity and
6:11:57interpretability is more pronounced than
6:12:00initially apparent the unused States in
6:12:03both models suggests that further
6:12:05optimization of number of states might
6:12:07be
6:12:09beneficial the high Reliance on volume
6:12:11change in both models indicates this
6:12:13feature is crucial for bitcoin price
6:12:15Dynamics the very low perplexity in both
6:12:18models suggests high confidence but this
6:12:21should be verified with additional out
6:12:22of sample testing to ensure it's not
6:12:25overfitting in conclusion while the 24
6:12:29State model shows better statistical fit
6:12:32the seven State models higher prediction
6:12:35accuracy and stability make it a
6:12:37stronger Contender especially for
6:12:41practical applications where
6:12:43interpretability and short-term
6:12:45predictions are important the choice
6:12:47between them would depend on specific
6:12:49use case and whether the additional
6:12:51complexity of the set 24 State model
6:12:54provides actionable insights for your
6:12:57Bitcoin analysis or trading strategy in
6:13:00conclusion while the 24 State model
6:13:02shows better statistical fit the seven
6:13:05State model higher prediction accuracy
6:13:07and stability make it a stronger
6:13:09Contender especially for practical
6:13:12applications where inter predicability
6:13:14and short-term predictions are important
6:13:18the choice between them would depend on
6:13:21the specific use case and whether the
6:13:23additional complexity of the 24 State
6:13:25model provides actionable insights for
6:13:27your Bitcoin analysis or trading
6:13:31strategy all right so that's interesting
6:13:36so rip
6:13:3924 much love my
6:13:43brother we miss
6:13:47you but I think we got to go with
6:13:50seven we got to hit the Jaylen Brown on
6:13:53them I think we got to hit the Jaylen
6:13:55Brown on them all right so that's all
6:13:58this is saying I mean not entirely
6:14:03but okay the top two the top two are 24
6:14:08and S but it
6:14:12may be
6:14:15beneficial
6:14:17beneficial to use just the
6:14:23um just the seven so we can read States
6:14:33easier all right
6:14:36so that's that people are getting licked
6:14:38over here what's going on let's check it
6:14:41out oh yeah we had that huge Unis swap
6:14:44liquidation
6:14:46of 1
6:14:49million that was earlier
6:14:53today
6:14:58um looks like BTC is
6:15:01up to this liquidation here
6:15:06I'm sure we have it what day was this
6:15:08August 20th
6:15:10at
6:15:2414400 probably this right
6:15:29here yeah that's what I would guess it's
6:15:32my best guess but
6:15:42not much liquidity
6:15:43huh not much liquidation going on for
6:15:47this move right here to be real with you
6:15:50cuz what is this two hours of time so
6:15:52I'm looking for like a group of two
6:15:54hours of liquidation 7 to 11 that's 4
6:15:56hours dude this is it right here 9:30 to
6:16:0011
6:16:06so that only that move right here what
6:16:09does that tell you dog I don't know I
6:16:12don't know but that move right there 3%
6:16:16move was this much in
6:16:22liquidations it's
6:16:24interesting what does it tell you though
6:16:26I don't
6:16:27know let's go look at the seven regimes
6:16:30it's funny that we are back on model
6:16:32number two after we tested 82 models not
6:16:3682
6:16:44but I want to go ahead
6:16:52and I want to go get this back test
6:16:55going again I know I
6:16:58just asked
6:17:00AI for some sort of back test but I
6:17:02don't think I went through with it so
6:17:05multi- regimes here and then BT I want
6:17:08to go see what this back test looking
6:17:10like dude copy why did I copy it I don't
6:17:14know save close save
6:17:17close close close close close close
6:17:20close all right we closed them all and
6:17:23now I'm going to look at the back test
6:17:24because I want to see what the back test
6:17:26was
6:17:27doing was this the
6:17:30seven 34 weeks train scaler model 2 okay
6:17:34okay okay okay so I want to
6:17:38say I don't want to say anything yet but
6:17:41I want to run this model two again and
6:17:44then try to
6:17:47identify some of the yeah that's what we
6:17:49F to do right
6:17:51now I want to identify some of the
6:17:56um the
6:18:01things identify the seven regimes and
6:18:07label
6:18:17them
6:18:20um okay that's still a lot though but
6:18:23maybe I can like zoom in or
6:18:29something this gets a little bit of
6:18:31everything right still be
6:18:35oh I'll do that I'll do that I got these
6:18:38seven regimes right
6:18:51here dagn where I put those
6:18:55suckers right here we got them right
6:18:58here so this is
6:19:03um 24
6:19:05seven do you see seven in here I don't
6:19:07see it no
6:19:09[Music]
6:19:21volume yeah it's just that
6:19:24one okay
6:19:28so it's not what I wanted
6:19:31though sheesh
6:19:37I think it's this right
6:19:45here let's double check it you can see
6:19:48here that says bullish trending bearish
6:19:52trending so sideways consolidation I see
6:19:54it bearish
6:19:56trending I see it sideways consolidation
6:19:59I see it bearish trending I see it
6:20:11sideways consolidation I see it bearish
6:20:14trending I see it sideways consolidation
6:20:17I see it downward consolidation I see
6:20:21it what's the objective
6:20:32here I wish they stayed longer
6:20:38but I guess that's what the uh like
6:20:40three states does let's go look at the
6:20:42three states then
6:20:54dude how are we back to three
6:21:00states open reveal and finder
6:21:21okay why is this like this though that's
6:21:24weird because it says green
6:21:28here
6:21:30H let's check out three states here with
6:21:33um
6:21:36less frenzy risk on frenzy risk on
6:21:39frenzy risk on frenzy risk on
6:21:44[Music]
6:21:52[Music]
6:22:10H this only says risk
6:22:22on I'm not sure about
6:22:31this how do it only show Once
6:22:36this entire
6:22:38data from 2018
6:23:07and just keep switching this is a good
6:23:09exercise
6:23:11though strong neutral bullish it's just
6:23:14looping through these
6:23:18dude strong bear strong
6:23:22bear neutral
6:23:25bullish but these again these aren't
6:23:28these aren't real the names aren't real
6:23:30so it keeps tricking me in my mind
6:23:33they're not real
6:23:51all right I'm going to say build a back
6:23:53test here what are the
6:23:55[Music]
6:24:00uh the state
6:24:02uses zero
6:24:05two four and
6:24:22six okay so I have these
6:24:31states use the below as a template back
6:24:38test in order to send me back a fullback
6:24:45test
6:24:47code
6:24:54that
6:24:55test all possible buy and sell
6:25:01combos like
6:25:05buy at zero sell
6:25:10at next
6:25:14zero then
6:25:16test then test
6:25:19sell at two then four then
6:25:25six all different
6:25:28iterations of this back test and I think
6:25:32this is different what I did before
6:25:35It's tricky to for my brand to get
6:25:37around it but we'll
6:25:39see um I optimized it before but I was
6:25:43trying let me just get this out we'll
6:25:45see um all different iterations on this
6:25:48back
6:25:50test so after
6:25:55so test buying on two selling on zero
6:26:01then two then four
6:26:06Etc below is the
6:26:09model we will use get
6:26:14them all right so
6:26:17this we like deep up in here we're so
6:26:24deep copy
6:26:30path okay and then model two here scaler
6:26:39two
6:26:42okay by the way what is the model
6:26:472. job lib verse the
6:26:51scaler
6:26:54to okay let's paste in the BT code the
6:26:58BT code okay uh BT the back has code
6:27:03somewhere
6:27:06this is so intimidating bro I tell you
6:27:09that much so I'm
6:27:11sorry I'm really
6:27:20sorry send back pull
6:27:22code okay there we go and um let's watch
6:27:27her
6:27:28work trying to cut me off crazy crazy
6:27:32stuff hey man can I make a suggestion
6:27:34for this CLA rate limit issue please
6:27:36please please do please please please
6:27:40Josiah says oh I already WR that one go
6:27:42for it subp Jane hope you ate on time
6:27:48today uh not at all brother thank you
6:27:50for all you do my pleasure bro my
6:27:55pleasure agnan G says I've been trailing
6:28:00this VC Fork code Fork cursor AI that
6:28:04implements its own C- pilot you get 500
6:28:06CLA uses per month and cursor expands
6:28:09the context to your entire code base
6:28:13it's like 20 a month and if you have an
6:28:15anthropic API key different than the
6:28:18base Cloud plan you can use
6:28:20Claud at Cost
6:28:27instead I've been trailing this VC code
6:28:31Fork what do you mean no way so you're
6:28:34just watching okay you saw somebody get
6:28:36funded and you've watching their code
6:28:38that's
6:28:39dope smart smart man that imple that
6:28:43implements its own co-pilot you get 500
6:28:45claw uses per month and cursor expands
6:28:48the context to your entire code base
6:28:51it's like a $20 a month or if you have
6:28:54an anthropic API different than base you
6:28:56can
6:28:58um use claw
6:29:01instead dang
6:29:08I'm G write that
6:29:10down appreciate you sharing that how did
6:29:12you get started I just started on
6:29:14YouTube and just started learning it's
6:29:16all there um Josiah Valentine says don't
6:29:19want to distract you from the session
6:29:21here's a link that runs through how to
6:29:22leverage multiple models to complete
6:29:25task you never have to shut up bro stop
6:29:28it how do you get started how's it going
6:29:31I'm doing good man thank you was thanks
6:29:35do Uber scalping in one minute candle um
6:29:40something about scalping the one minute
6:29:41I'm not sure what do you think about SC
6:29:43in one minute it's harder down there
6:29:45Zeus what's up bro 777 much love hope
6:29:48you're doing great he said thank you 777
6:29:51ditto Jane yo ditto even ditto
6:29:54ditto Zeus you you on Discord Discord
6:29:58question Mar we have a Discord Zeus
6:30:01please one sec finally
6:30:04I've joined the Discord zo says by the
6:30:06way I'm working on a project for a SAS
6:30:08course at my University I am extremely
6:30:10impressed with the Claud Dev plugin for
6:30:12vs code it's like Claw on
6:30:16steroids that's the second time I've
6:30:18heard that that's the second time I've
6:30:21heard
6:30:31it Claude Dev
6:30:45e
6:30:53eest appreciate your shares
6:30:58Fellers okay so thank you thank you both
6:31:01for that
6:31:16plug in by sa
6:31:19rzan there a few YouTube videos on it if
6:31:22you search something like claw Dev that
6:31:24might help at first to get a hang of it
6:31:27I gotta go peace see you Jane 777
6:31:38to be honest they got you for this 20
6:31:41this it's super
6:31:51valuable nice I'll just check out the
6:31:54YouTube videos or
6:32:02something why YouTube
6:32:05video okay there we
6:32:13go it's called cursor AI
6:32:18essentially essentially actually for
6:32:30real so test States here 024 four and
6:32:34six by stay
6:32:36okay let's see how this do do
6:32:41dude seven regimes multi- regimes o o
6:32:46testing let's call these
6:32:49uh
6:32:57testing where that back test at you come
6:32:59up in here the testing dude testing
6:33:07okay let's add it
6:33:10um what do we do here we say this is uh
6:33:22BT
6:33:24seven states I don't know
6:33:27dog I don't
6:33:30know I know nothing
6:33:35except I'mma keep
6:33:38going I'm keep going and
6:33:41going and then I'm going keep
6:33:44going I'm going get
6:33:46rxed I'mma keep
6:33:49going I'm going be stuck I'mma keep
6:33:52going every day
6:33:55dude every
6:33:57day all right look at
6:34:02that uh-oh
6:34:05what did I just show
6:34:09you what did I just show you dude win
6:34:12rate 60% 1.1 sharp return
6:34:1823 okay I
6:34:21mean exposure times 11% that's what I
6:34:24just showed you is that worth it to you
6:34:26I like
6:34:27it I would give up a few percentage
6:34:30points for that
6:34:46okay let's go ahead and copy this error
6:34:52here man I love this game
6:34:55dog love this game so much it's so
6:35:01hard I love it so much it's so hard
6:35:06boohoo let's get it every day dog come
6:35:10on come
6:35:12on that's just a warning cousin that's
6:35:15great to see let's see what it said
6:35:19though those warnings are related to the
6:35:22Boke Library used for plotting back
6:35:24testing package they don't affect the
6:35:27functionality of your back test but
6:35:29let's address them to clean up the
6:35:31output clean up the output for me
6:35:35please clean up the output for me please
6:35:38cuz you know we're just getting started
6:35:40Oh I thought we've been going a while
6:35:42now no way no way dude why why would I
6:35:47slow down I'm sorry I'm sorry every time
6:35:49you see me here I'm working on something
6:35:51new my
6:35:55bad my
6:36:00bad I'mma keep going
6:36:06I'll show you everything
6:36:08though I'll show you everything
6:36:12bro every
6:36:15day what I find you find that's the
6:36:19game replace the original plotting call
6:36:22with this
6:36:23okay let's go to the plot plot plot
6:36:28plot BT plot oh is that the one
6:36:36nah yeah it is all right all right let's
6:36:39go ahead and run in
6:36:42now Warning free beautiful beautiful
6:36:48beautiful okay
6:36:56dude what else can I
6:36:59do there's a lot I can do
6:37:13I like this a lot
6:37:43all right so I'm just going to save this
6:37:50then
6:37:53dog stop playing stop playing dog that's
6:37:57out of sample data
6:37:59too I like this dude 23% return is not
6:38:04better than Buy and Hold but you're only
6:38:06holding 12% of the time
6:38:09dog 8 months 9 months 8
6:38:13months stop it stop it stop playing with
6:38:17your boy I don't know how many hours
6:38:19this has been dog but it's been a lot of
6:38:22hours been here a long
6:38:26time feel like this is just a start
6:38:29though
6:38:31[Music]
6:38:36Max trade time two
6:38:38days expectancy how L is your boy
6:38:41[Music]
6:38:44Nate 2.3 okay this doesn't mean anything
6:38:47though all it means is we're just
6:38:49getting
6:38:51started seven states let's do 24 States
6:38:55cousin I got mad I got mad things I need
6:38:58to do now okay live Ops copy path copy
6:39:02path copy path Cy relative path read me
6:39:05live Ops where you at Dog there you are
6:39:09okay so I'm just going to keep vibing
6:39:10out here until I'm done I don't know
6:39:12when I'm done I'm never going to be done
6:39:24dog I'm never going to be done there is
6:39:27no done in this game that's why I love
6:39:32it hey
6:39:34mm this is a great looking back test
6:39:41with 11%
6:39:46exposure and 1.1 sharp it's not the
6:39:50highest but you know it's a good
6:39:53start now I want to
6:39:57test with uh eth hourly data let's grab
6:40:01that eth data right now
6:40:04let's just grab it right now eth data
6:40:06how many weeks we want is the
6:40:08question do I already got
6:40:11it no I'm going to get it I'm G to get
6:40:13it from here how many weeks we want is
6:40:15the question let's do 100 weeks let's do
6:40:17104 weeks let's do 104 weeks that's two
6:40:21years
6:40:22dog maybe I should go more
6:40:25though 10 weeks what do I got in here
6:40:28already I got 34 weeks I got a th weeks
6:40:31of BTC data okay I should try it on that
6:40:34for
6:40:36sure copy
6:40:39path this we probably just break it but
6:40:42it's
6:40:44okay there's a th
6:40:47weeks of data and I'mma put this janky
6:40:50little comma there
6:40:52because no wait wait wait wait needs to
6:40:56be in the data Boulder one no janky
6:41:00comma no janky comma
6:41:06no comma no
6:41:20comma I'm just going to get all the data
6:41:23I can yeah I'm just going to do like 500
6:41:25weeks of e
6:41:27data I can't
6:41:29imagine it's going to disrupt things but
6:41:33what's going to happen is it's going to
6:41:35not be as good I mean but it wasn't as
6:41:37good as buying hole anyways that's fine
6:41:39how many weeks let's do 200 weeks so
6:41:41that's like four years just I got to be
6:41:43conscious of the time that takes to
6:41:45download all the data but now it's
6:41:47downloading
6:41:48[Music]
6:41:50um one of the many things I show you in
6:41:53the boot camp the boot camp shows you
6:41:55step by step how to automate your
6:41:56trading 100% money back guarantee always
6:41:58always
6:42:01always if if you got questions about any
6:42:04of this stuff it's probably not because
6:42:05you can't do it it's just you haven't
6:42:07started yet so this will help you out a
6:42:09lot I wish I had this when I got started
6:42:12step by step how to automate your
6:42:13trading how to back test how to build
6:42:16your own Edge and um yeah I'm flying bro
6:42:19I'm
6:42:20flying I'm out of here I'm gonna keep
6:42:26going all right so
6:42:34see what happens can I just pop this in
6:42:36let's see if I can just pop it
6:42:37in yeah this is out of sample data copy
6:42:40relative path
6:42:45here and back test seven states this
6:42:50is for that okay and now we'll try it
6:42:55with this
6:42:57thousand I don't expect it to be better
6:43:00than Buy and Hold because this this is
6:43:03all Bitcoin data and there's uh that's a
6:43:06pretty good strategy in retrospect to
6:43:08just buy and
6:43:10hold from the start of
6:43:13Bitcoin but that's not what we're
6:43:15looking
6:43:17at all right so copy path okay I'm just
6:43:22going to plop in that data
6:43:24now let's just say uh let's mark this
6:43:27one out and then maybe my AI can just do
6:43:29it for me yes thanks shorty look at that
6:43:31I don't even have to code I just move
6:43:33things around playing Legos all right
6:43:36let's see if it can do it with a th then
6:43:39it looks like it's working so I bet we
6:43:42have our information here in a little
6:43:44bit of time there it is dude okay how's
6:43:48that look to
6:43:50you I don't know it's 10 years of
6:43:55data 2015 wowers bro cool stuff cool
6:44:01stuff
6:44:03all right so the profit factor is 1.6
6:44:06the expectancy is
6:44:091.83 sqn is
6:44:111.47 280 trades sharp ratio kind of low
6:44:16the return 56%
6:44:22[Music]
6:44:24annualized but like I said it's the Buy
6:44:26and Hold is
6:44:27like obviously goat here exposure time
6:44:3268% % so over 3,000
6:44:41days it's not the
6:44:53goat but you can't go back 3,000 days in
6:44:57Bitcoin that's the that's the tricky
6:44:58part about back testing against a lot of
6:45:01Bitcoin data
6:45:05so I like to look at other things like
6:45:07the
6:45:08expectancy that's
6:45:12nice The Profit factor is
6:45:16nice win rate is
6:45:28solid draw down is not solid
6:45:39but it's you
6:45:40know that return
6:45:43[Music]
6:45:46um oh cool I got the uh eth
6:45:57data I'm going to do the same thing for
6:45:59uh soul but all of these are just up
6:46:01over the years which which is makes it
6:46:04tricky tricky tricky I'm going to do it
6:46:05for sa as well so let's go ahead here
6:46:10and try it
6:46:23out eth 1
6:46:27hour okay this is the 1 hour data for
6:46:29200 weeks it
6:46:31says and I haven't tried a different
6:46:38um choer him out of here get get them
6:46:41ghost it's time to go home buddy quit
6:46:44the game if you're going to play the
6:46:46game the same way you always play quit
6:46:47the game you're going to get liquidated
6:46:50over and over and over and over again
6:46:52till you can't even play the game no
6:46:54more so quit the game now put the game
6:47:01down put the game
6:47:03down dude come on stop
6:47:07playing stop
6:47:10playing stop playing like
6:47:15that
6:47:16so you can see it's a 380 per return I
6:47:21like this one though exposure is only
6:47:2433% so to get that close of return
6:47:27versus Buy and Hold that's like a lot of
6:47:29capital free that's a lot less risk
6:47:33Sharp's not great
6:47:3848
6:47:40228 this is cool though this is a cool
6:47:43new
6:47:46um new back test here
6:47:54so yeah that's pretty neat it's pretty
6:48:01neat it's got me a lot it got me
6:48:03thinking a lot like
6:48:05um where' it go from here with
6:48:10this cuz this back test is really just
6:48:12like a
6:48:14filter it's a regime
6:48:18Creator it's a regime
6:48:25Creator
6:48:27so I'm going to grab this one as well
6:48:38I'll just leave this here for
6:48:40now 23% Buy and Hold
6:48:4335 this is
6:48:45a I
6:48:48mean as a filter and now layer
6:48:51strategies in could be
6:48:54interesting we'll
6:48:57see we shall see
6:49:19copy this down here
6:49:46I already did that dum dummy come on
6:49:49baby boy come
6:49:54on
6:49:55copy relative
6:49:58path okay so that's that and this is
6:50:04not as good return as Buy and
6:50:08Hold if you would have bought on
6:50:111023 of this
6:50:13date to this date four
6:50:17years but a lot of free
6:50:27capital and this is just a
6:50:29filter so I'm actually super interested
6:50:32in this now
6:50:35man I need to write some stuff
6:50:47down
6:50:50so eth
6:50:52was like
6:50:55four 380 return verse 500 Buy and Hold
6:51:02with 30 3% whole
6:51:04time but that's pretty biased like I
6:51:07just wrote that down there's another one
6:51:08in there um what I look forward to what
6:51:12I look forward to here is that these are
6:51:19simply
6:51:23regimes or
6:51:27filters so being able to find a
6:51:31profitable in the past not guaranteed
6:51:34the Prof in the
6:51:35future
6:51:39Strat off regime
6:51:42changes is dope but also the market just
6:51:45goes straight up so I'd be interested to
6:51:47see how this looks on other data yo
6:51:51rigar what's up
6:51:52dude Samir says is the strategy finding
6:51:55the hidden states on the entire data set
6:51:58once or is hmn predicting future States
6:52:01based on a window of past data
6:52:04[Music]
6:52:06um I uh it's it's past data past data
6:52:10it's trained on past data and predicts
6:52:13with that
6:52:14model Jamie says hey man I see your
6:52:16streams appear on my
6:52:18YouTube what is it that you're doing hey
6:52:23dude
6:52:25welcome much love to
6:52:28you I am uh I'm coding I'm coding a uh
6:52:32machine learning model it's
6:52:35a hidden markof
6:52:38model and we've been playing around with
6:52:40it for a
6:52:41while and we got the back testing it
6:52:43today it's a pretty big uh pretty big
6:52:46moment actually pretty big
6:52:51moment but I still think it's just a
6:52:54start I still think it's just the
6:52:59start cuz there's so many ways we can
6:53:01slice and dice this
6:53:05now so live Ops I wanted to write that
6:53:13okay I want to do the same
6:53:17thing I want to do the same concept
6:53:27here for the
6:53:3024 um
6:53:34State model I want to do the same type
6:53:39of back test where we test all
6:53:45variables
6:53:47again uh
6:53:50but for
6:53:54this these
6:53:56many more states listed above
6:54:06which appear most in the 24 model okay
6:54:13and then we got to get that code because
6:54:14I don't know if she remembers I don't
6:54:18even remember so you know how's she F to
6:54:22remember if I can't even remember cuz
6:54:23she's smarter than you
6:54:31Moon she can definitely remember that
6:54:34she can defin I didn't know it was the
6:54:36last one it's just been a while we've
6:54:38been over here
6:54:40executing what a great teammate she is
6:54:43huh it's crazy it's crazy stuff here
6:54:46dude trained model let's go to 24 copy
6:54:51path she probably doesn't even need this
6:54:53to be honest she's so smart
6:55:12send back full
6:55:31code
6:56:01for for
6:56:55what
6:57:01up
6:57:31for for
6:59:01e
6:59:31e e
7:00:13all right
7:00:14so this is the 24 back
7:00:17test just curious you
7:00:22know 24 rest in peace
7:00:26King all
7:00:29right R it
7:00:37all right so looks
7:00:45uh oh this is on the that data too wait
7:00:49what dat is on worst
7:00:51trade good expectancy good profit
7:00:57Factor beats the Buy and Hold
7:01:04uhoh what does that tell me
7:01:07dog that tells me
7:01:10good dang
7:01:14dude another
7:01:28one H so 240 days
7:01:49all right so that's that and what data
7:01:52was that on
7:01:53dude 31 we 34 weeks okay let's try on
7:01:58some other data then
7:02:05this is this one's almost impossible to
7:02:07beat but might as well try
7:02:11it the
7:02:13uh
7:02:15data from BTC 10
7:02:18years 10 year BTC data let's see it
7:02:22let's see it dude let's see you
7:02:27dance all right it's going to take a
7:02:30little bit of time
7:02:37there it
7:02:38is expan See's
7:02:49nuts 8,000% verse
7:02:53um 23,000 so no didn't beat
7:02:59it 53
7:03:0588 win rate is good though
7:03:1154% expectancy is good draw down how's
7:03:15that draw down looking 74% so not so hot
7:03:22but I think that's pretty
7:03:25standard or this 70% exposure time
7:03:31and volatile asset I don't know what to
7:03:34conclude I have no conclusion except low
7:03:38s s Lo
7:03:41s all
7:03:43right let's go
7:03:46ahead paste it here B it up
7:03:50here okay
7:03:53so it's cool it's really cool I like
7:03:56this idea I like this idea of cutting
7:03:59things up into regimes and having AI
7:04:03mess with
7:04:05it so here I'm going to say this is the
7:04:08out of sample out of
7:04:11sample data here this is the first one
7:04:16okay and
7:04:18now this one here is the BTC 1
7:04:26hour all right sick so let's try that
7:04:30eth here let's say copy
7:04:34path um I don't know why I'm up in there
7:04:37but copy path
7:04:40and say data path let's go ahead and
7:04:42delete this one be out dog data path
7:04:46okay that's cool with me hey guess what
7:04:49else we got some more data too so oh you
7:04:53tickling me oh you're hungry again you
7:04:55hungry I'm hungry too we both hungry
7:04:58shorty we both hungry I'mma feed you
7:05:01right now
7:05:02you grow and grow and
7:05:06grow um you stay on track just for a few
7:05:09more seconds here and then I'm G get you
7:05:15something I want to get this Soul one
7:05:18hour in
7:05:20here all right I got it in
7:05:23there um this here
7:05:26is the detailed results for that Strat
7:05:3124
7:05:35States this one was for the eth okay it
7:05:39looks solid
7:05:422.54 sharp a little better 7 beats the
7:05:45Buy and Hold it's cool to
7:05:49see exposure time is 2/3 of the
7:05:54time
7:05:56okay
7:05:59so what was that that was the e 1 hour I
7:06:05believe let's try something new dude
7:06:07let's try something
7:06:10new let's say data
7:06:14path
7:06:16equals data path data path data path
7:06:19data path data path data path data path
7:06:23copy path copy
7:06:26path paste it in okay now we're going to
7:06:30say up here okay
7:06:33copy run
7:06:35it all right that one's going to take a
7:06:37minute but this is it this is that oh
7:06:40snap what is this
7:06:42though okay
7:06:47so this was eth copy relative path here
7:06:51ET States I just want to make sure that
7:06:53it's here so I know what I'm doing data
7:06:56okay what did the what did that chart
7:06:58looked kind of
7:07:07so this is the draw Down August
7:07:1020
7:07:1622
7:07:19interesting profit factor is 10 only six
7:07:23trades
7:07:27interesting and it beat b and hold
7:07:31but not very many trades
7:07:36so you know so quick look six trades
7:07:39this one had 150 trades a little more
7:07:41statistically significant better than
7:07:43Buy and Hold but then the BTC one from
7:07:4610 years is
7:07:50not what about BTC for
7:07:57like I don't have it I could get it
7:08:00pretty easily though
7:08:02I could do other time frames but you
7:08:04know I think I don't want to over test
7:08:06this just it is what it is it's good
7:08:10information to have what you going to do
7:08:13with that information
7:08:14though that's a cute
7:08:17dude that's the
7:08:20queue
7:08:27so this is a good place this is a good
7:08:29place to now look at this and see how do
7:08:33we layer
7:08:34it how do we take it from
7:08:37here I'm always constantly just
7:08:39following this if you're ever wondering
7:08:41what I'm doing it's
7:08:43this every day every day for the next 60
7:08:46Years bro RBI system for AO trading it's
7:08:49just my simplified version of this the
7:08:53process of automating your trading comes
7:08:54down to researching trading
7:08:57strategies watching videos listening to
7:09:00podcasts
7:09:01reading
7:09:03books I didn't love reading growing up
7:09:05but now you're just
7:09:07searching searching for ideas one idea
7:09:09can change the
7:09:11world then back test those ideas
7:09:14strategies to see if they actually work
7:09:16in the past so the idea today was the
7:09:20hmm a markof model to predict the
7:09:24different states in the market and then
7:09:26have a back test that trades based off
7:09:28of the changes of those States
7:09:35see if it works in past data and then if
7:09:39that's Prof profitable in the past it
7:09:41might work in the future so I always
7:09:43start with tiny size but most of the
7:09:46work is in the r and the B and everybody
7:09:49wants to build a bot including myself
7:09:53just take that lesson from my experience
7:09:56seriously and don't just go build a bot
7:09:58don't buy a bot on the internet none of
7:10:01that that dude
7:10:03please if everybody's running the same
7:10:05algo it's going to converge to zero
7:10:07profits over time pretty much guaranteed
7:10:09it's just
7:10:11math and I think we went over this a few
7:10:13times in
7:10:15this this uh this stretch uh this
7:10:19journey this Marathon this journey
7:10:21that's what it's been it's been a
7:10:29journey that's why machine Lear seems to
7:10:31be hard for trading because if everybody
7:10:33can predict the price then the price is
7:10:35not going to be the same anymore because
7:10:37everybody's going to trade off that
7:10:42price it's not like if everybody knows
7:10:45the prediction of the
7:10:48weather it's going to change the weather
7:10:54no it's
7:10:58different it's a fun Journey
7:11:02I like this journey a lot

About this transcript

This page contains the full transcript of how to actually use ai for trading by Moon Dev, generated from the public captions YouTube serves with the video. The transcript has 45,437 words across 7,507 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.