Workshop Real-time Adaptive Machine Learning Algorithms with Python

on YouTube ↗03h 01m

" summary

Workshop follows code from the GitHub chatterj0 repo and runs in Docker TensorFlow JupyterLab (requires ~25 GB). It implements streaming feature extraction: adaptive median with update wk+1 = wk + sign(xk−wk) to flag anomalies (thresholds 50–75% of median), adaptive PCA/whitening/LDA for representation and classification, and the ZOO algorithm to compute principal eigenvectors online. Experiments: UCI clickstream (165k rows) and Yahoo Webscope anomalies, MNIST reconstructed from 784→100 PCs (~7.8× reduction), and 128→16 compression; principal eigenvector converges quickly, secondaries show 100-sample ramp-up but steepest-descent reduces that to ~10 samples.

AI-written summary · about

" chapters (33)

1. 0:00Start
2. 0:18Clone the Adaptive Repo
3. 8:56Coupon Code
4. 10:45Introduction
5. 16:50Features
6. 18:55Examples of Feature Extraction
7. 21:55Principle Component Analysis
8. 22:54Covariance Matrix
9. 25:24Linear Discriminant Analysis
10. 26:25Lda Example
11. 50:19Adaptively Compute the Median
12. 50:40Median Algorithm
13. 1:00:07Example of Character Recognition
14. 1:06:02Download the Docker
15. 1:07:24Introduction of the Book
16. 1:18:44Feature Drift
17. 1:23:09Data Representation Features
18. 1:25:12Data Whitening
19. 1:26:37Inverse Square Root of the Correlation Matrix
20. 1:35:01Why Do We Want To Do It Adaptively
21. 1:42:15The Zoo Algorithm
22. 1:52:12Stationarity of Data
23. 1:55:59Adaptive Median Algorithm
24. 1:59:02Chapter 2 Here's the Anomaly Detection Example
25. 2:21:47Examples of Non-Stationary Data
26. 2:29:43Principle and Minor Eigenvector Computation
27. 2:29:57Example of Image Representation
28. 2:34:08Example Two
29. 2:36:29Steepest Descent
30. 2:45:25Gradual Drift
31. 2:51:59Principal Eigenvector
32. 2:56:59Compute the Generalized Eigenvectors
33. 3:00:40Summary

" related talks

• 45mReal-time, Streaming Advanced Analytics....Chris Fregly · Data By the Bay 2016
• 01h 17mOne workshop that data scientists don't want you to attend...Oliver Laslett · PyData London 2017
• 30mReconceptualizing Machine Learning for the Real-time WorldPatrick Deziel, Prema Roman · PyData Global 2023
• 01h 03mWorkshop: Intro to GPU data science with RAPIDSRhys Williams · Big Data & AI Toronto 2020
• 31mMonitoring machine learning models in productionArnaud Van Looveren · PyData Global 2020
• 39mAnd Then There Are AlgorithmsDanilo Poccia · Codemotion Rome 2018
• 02h 19mWorkshop: Demystifying Natural Language Processing using Python (Scikit-Learn/ Keras)Vaibhav Srivastav · PyCon APAC 2019
• 47mData Science and Machine Learning with PythonArgeo Tuble Alecha · PyCon APAC 2019