ML Regression - Linear regression, MLPs, DL nets, Linear Plus Non-Linear, XGBoost, ONNX (part 4) — Transcript

Full transcript

1. 0:10Okay, so we're recording.
2. 0:15All right. Okay, so to recap, right? So
3. 0:19today we did
4. 0:21regression models with the wine quality
5. 0:23data set. We ran
6. 0:25linear regression, multi-layer
7. 0:27perceptron, deep learning, linear plus
8. 0:30non-linear, and ONNX. Okay? Uh no,
9. 0:34sorry. And
10. 0:35XGBoost. Those were all the models.
11. 0:39And we measured performance. We saw that
12. 0:42deep learning had a score of 39 R
13. 0:45squared, and then XGBoost had a score of
14. 0:4944 R squared.
15. 0:51Okay? So now
16. 0:55Um then after that, we did look at
17. 0:58deploying
18. 0:59the model via ONNX, and I gave you the
19. 1:01code for that.
20. 1:03I showed you that the ONNX file then is
21. 1:06loaded on GitHub, and I showed you the
22. 1:09HTML and and CSS files.
23. 1:12There's a link in the homework
24. 1:18page.
25. 1:19Okay, what is this here?
26. 1:24Okay, let's go back, I guess, to
27. 1:31Okay.
28. 1:33Then we went to
29. 1:35uh the content page in Blackboard,
30. 1:40and I gave you the homework assignment.
31. 1:43The homework assignment requires you to
32. 1:45find a classification data set and a
33. 1:47regression regression data set of your
34. 1:49choosing,
35. 1:50and do the analysis as I did today for
36. 1:53for regression, and the analysis as I
37. 1:55did a a week ago for classification.
38. 1:58Once you have finished, and you need to
39. 2:00measure performance with the appropriate
40. 2:02metrics. Once you have finished you will
41. 2:04need to go to GitHub pages and deploy
42. 2:06the best model
43. 2:08um
44. 2:10using ONNX
45. 2:12and and you know, provide a link in your
46. 2:15GitHub pages so that you can run the
47. 2:18regression prediction or the
48. 2:20classification prediction with new
49. 2:23values.
50. 2:25Got it? That's the homework assignment.
51. 2:29All right, any questions on the
52. 2:30homework?
53. 2:32Okay, once that's completed then that's
54. 2:34good. Um the last part of this then
55. 2:38is to
56. 2:40if we go back to the code, right? We
57. 2:43the last thing we did is exporting the
58. 2:45models.
59. 2:46But there is one last component in the
60. 2:49Jupyter notebook that I was using.
61. 2:51If you notice, you know, we looked at
62. 2:54performance, we exported it.
63. 2:56We did XG boost.
64. 2:59We saw that XG boost does
65. 3:02um
66. 3:03really well. And then you might say, but
67. 3:06my XG boost model is the best.
68. 3:10Do you see that?
69. 3:11So if your XG boost model is the best
70. 3:15then you may also want to export that.
71. 3:18Okay? So here's a simple example. I'm
72. 3:21not going to make this a requirement,
73. 3:23but I'm giving you the code for those of
74. 3:25you that want to do it, go ahead and
75. 3:28export the XG boost model as well. Okay?
76. 3:32So I'm only asking for the like ONNX as
77. 3:34you saw in the description.
78. 3:36Um
79. 3:39See? Using ONNX, using ONNX, but
80. 3:42feel free to give this one a shot. It's
81. 3:44not hard to do. It's just more work.
82. 3:48The ONNX and GitHub so deploy XG boost
83. 3:51model. So you can see the code is is
84. 3:54here. It's pretty much the same thing.
85. 3:56Set up like the size of it.
86. 3:58On it, you know, the the
87. 4:00convert XG boost ONNX models, save the
88. 4:03model,
89. 4:05and
90. 4:06um
91. 4:07you would save the model as this XGBoost
92. 4:10wine quality ORT. So, it's just these
93. 4:12two lines of code. Okay?
94. 4:15So, if you want to try that, go ahead,
95. 4:17but I'm not making it a requirement. Got
96. 4:20it?
97. 4:22All right. Um so, that's the code for
98. 4:24that.
99. 4:26So, I I leave it to you guys to look
100. 4:29through
101. 4:30and then you would have to deploy it,
102. 4:32obviously, the same way on GitHub if you
103. 4:35choose to do that.
104. 4:38And that's it. That's the Jupyter
105. 4:40notebook. So, essentially, I'm saying it
106. 4:43is possible to deploy XGBoost. It's not
107. 4:45hard, but I'm not requiring.
108. 4:49Any questions?
109. 4:52No questions? Ev- Everything super
110. 4:54clear?
111. 4:55All right, great. All right. So, yeah.
112. 4:57So, I think this is actually a good
113. 4:59stopping point. Um
114. 5:02I don't think I missed anything. We
115. 5:04finished this topic.
116. 5:05Uh next week, as I said, we have
117. 5:08um exam two.
118. 5:11Don't forget. Please, do not miss exam
119. 5:13two. Uh the first half, I'll probably
120. 5:15introduce neural input optimization
121. 5:18as that's going to be the topic of your
122. 5:21term project.
123. 5:22I do I will make it a requirement for
124. 5:24you to deploy the NIO model similar to
125. 5:27this, but instead of ONNX, we will use
126. 5:30the tool called Fast API for that one,
127. 5:32and that'll make sense later why we need
128. 5:34to do that. And then after that, we'll
129. 5:36move on to like gradients and things
130. 5:38like that, more into the background of
131. 5:40machine learning,
132. 5:42deeper. All right. Yeah, so this is a
133. 5:44good stopping point. I'll stay for a few
134. 5:46minutes if you got have guys have
135. 5:48questions, and please don't forget the
136. 5:50exam.