Rust + Jesse = Match made in heaven — Transcript

Full transcript

1. 0:00Hey guys, it's Al. This release is all
2. 0:02about speed. That means faster back
3. 0:04tests, optimization runs, and soon it's
4. 0:06going to be faster Monte Carlo
5. 0:08simulations and also machine learning
6. 0:10training. Yes, I'm spoiling a little bit
7. 0:11from the future, but it's coming. Now,
8. 0:13the main reason behind these faster back
9. 0:15tests are first this refactor that I did
10. 0:18for how we handle candles in back
11. 0:20testing. And to give you an example,
12. 0:22this is this very heavy back test that I
13. 0:24had for 3.4 years. And it's also using
14. 0:27the one minute time frame. It's very
15. 0:29heavy. It's also using bigger time
16. 0:31frames. And if you take a look at here,
17. 0:33it took us 625 seconds to finish this
18. 0:36back test. But after these refactors, I
19. 0:38finished the exact same back test in
20. 0:40only 138 seconds. So that is almost six
21. 0:44times faster right there. And notice
22. 0:46that all the numbers are exactly the
23. 0:48same. So that's number one. If your
24. 0:50strategy is also using multiple time
25. 0:52frames, you're going to see
26. 0:53improvements. Now how much of it depends
27. 0:55on the duration of your back test and
28. 0:57the exact time frames that you were
29. 0:59using. Now the second reason why this
30. 1:00release is faster is because of the
31. 1:02integration of the Rust language. Now if
32. 1:04you don't already know Rust, it's this
33. 1:06very shiny language that these is all
34. 1:08the apps are using in order to improve
35. 1:10the performance. Now instead of
36. 1:11rewriting the entire Jesse code with
37. 1:13Rust which is going to make it very
38. 1:15unreadable and hard to maintain, I only
39. 1:17implemented the parts that were critical
40. 1:19for performance. Now that part is still
41. 1:21ongoing because we were already
42. 1:23optimized using C libraries such as
43. 1:25numpy or pandas. But wherever that's not
44. 1:27the case, I am going to write it in rust
45. 1:29from now on. But starting now, I
46. 1:32implemented many of our indicators with
47. 1:34this language. Now if you remember just
48. 1:36a few months ago, we stopped using tool
49. 1:39as a dependency of Jesse. And the reason
50. 1:41was because it made it very difficult
51. 1:43for people to install Jesse in the first
52. 1:45place because that library wasn't being
53. 1:47maintained as well as I needed it to.
54. 1:49But now the new implementation that I
55. 1:51did not only is faster than what we had
56. 1:53just few weeks ago, but it's even faster
57. 1:55than the toll library, which means I
58. 1:57guess I could claim that Jesse is now
59. 1:59the fastest indicator library out there.
60. 2:01I don't know. I haven't benchmarked like
61. 2:03all of them, but I'm pretty sure it is
62. 2:05super fast. Now, the code for this
63. 2:08integration that I did is open source
64. 2:09and you can find it on our repository
65. 2:12under Jesse Rust, which doesn't concern
66. 2:14most of you, but if you are curious how
67. 2:15I did it, you can check it out. But now
68. 2:17let me show you how much faster it got.
69. 2:19So remember this back test that I said
70. 2:21is like six times faster now and it took
71. 2:24138 seconds. Well after the new Rust
72. 2:27implementation, we are finishing the
73. 2:29same back test with the exact same
74. 2:31numbers in only 95 seconds. So that is
75. 2:34almost 40% faster just because of the
76. 2:37new indicators. May not really care much
77. 2:40about 40% faster if your back test is
78. 2:42very simple and you're just running it a
79. 2:44few times. But imagine this that if you
80. 2:46were running the optimization mode for
81. 2:48the same back test, if you were spending
82. 2:501 hour, now you're going to finish the
83. 2:52same thing in 36 minutes. But if it's
84. 2:55not 1 hour, if it was 4 hours or 10
85. 2:57hours, you get the idea. It's going to
86. 2:59add up very quickly. And especially in
87. 3:01the upcoming Monte Carlo simulations,
88. 3:03this is going to be a huge win. In fact,
89. 3:05that's why I'm tackling this right now
90. 3:07because I know the value of it for both
91. 3:09Monte Carlo and AI training such as
92. 3:12machine learning. and I couldn't be more
93. 3:13excited about those. Now, I also did a
94. 3:15lot of improvements and fixes in this
95. 3:18release. So, if you haven't checked them
96. 3:19out, make sure to check the change log
97. 3:21page on our documentation. Thank you so
98. 3:22much for supporting the project. As
99. 3:24always, I'll see you soon.
100. 3:26[Music]
101. 3:35[Applause]
102. 3:39[Music]