Machine Learning with H2O.ai - Budapest Data Science Meetup - July 2016

Transcript

1. Machine Learning with H2O.ai Szilárd Pafka, PhD Chief Scientist, Epoch Budapest Data Science Meetup July 2016
3. Disclaimer: I am not representing my employer (Epoch) in this talk I cannot confirm nor deny if Epoch is using any of the methods, tools, results etc. mentioned in this talk
10. Supervised Learning y = f(x) train: “learn” f from data X (n*p), y (n) score: f(x’)
11. Supervised Learning y = f(x) train: “learn” f from data X (n*p), y (n) score: f(x’) algos: k-NN, LR, NB, RF, GBM, SVM, NN, DL… goal: max accuracy measure (on new data)
12. Supervised Learning y = f(x) train: “learn” f from data X (n*p), y (n) score: f(x’) algos: k-NN, LR, NB, RF, GBM, SVM, NN, DL… goal: max accuracy measure (on new data) f ∈ F(θ) min θ ( L(y, f(x,θ)) + R(θ) ) on train set
13. Supervised Learning y = f(x) train: “learn” f from data X (n*p), y (n) score: f(x’) algos: k-NN, LR, NB, RF, GBM, SVM, NN, DL… goal: max accuracy measure (on new data) f ∈ F(θ) min θ ( L(y, f(x,θ)) + R(θ) ) on train set evaluate on separate test set /cross validation
14. Structure/Hyperparameters λ min θ ( L(y, f(x,θ[,λ])) + R(θ,λ) ) often λ ~ capacity/complexity
15. Structure/Hyperparameters λ min θ ( L(y, f(x,θ[,λ])) + R(θ,λ) ) often λ ~ capacity/complexity
16. Model selection: Need Vary λ and get model with best accuracy on validation set
17. Model selection: Need Vary λ and get model with best accuracy on validation set Evaluate final model on test set /cross validation
18. overfitting
19. http://datascience.la/meetup-summary-winning-data-science-competitions/
21. syllabus, slides, book readings, code, paper refs, sample project
22. http://www.cs.cornell.edu/~alexn/papers/empirical.icml06.pdf http://lowrank.net/nikos/pubs/empirical.pdf
23. http://www.cs.cornell.edu/~alexn/papers/empirical.icml06.pdf http://lowrank.net/nikos/pubs/empirical.pdf
24. http://datascience.la/meetup-summary-winning-data-science-competitions/
26. data size [M] training time [s] 10x Gradient Boosting Machines
28. Disclaimer: I’m not affiliated with H2O.ai. It’s just that in my opinion H2O is a machine learning tool with several advantages. There are many other good tools (and many more awful ones).
29. - high-performance implementation of best algos (RF, GBM, NN etc.) - R, Python etc. interfaces, easy to use API
30. - high-performance implementation of best algos (RF, GBM, NN etc.) - R, Python etc. interfaces, easy to use API - open source - advisors: Hastie, Tibshirani
31. - high-performance implementation of best algos (RF, GBM, NN etc.) - R, Python etc. interfaces, easy to use API - open source - advisors: Hastie, Tibshirani - Java, but C-style memalloc, by Java gurus - distributed, “big data”
32. - high-performance implementation of best algos (RF, GBM, NN etc.) - R, Python etc. interfaces, easy to use API - open source - advisors: Hastie, Tibshirani - Java, but C-style memalloc, by Java gurus - distributed, “big data” - many knobs/tuning, model evaluation, cross validation, model selection (hyperparameter search)
33. - high-performance implementation of best algos (RF, GBM, NN etc.) - R, Python etc. interfaces, easy to use API - open source - advisors: Hastie, Tibshirani - Java, but C-style memalloc, by Java gurus - distributed, “big data” - many knobs/tuning, model evaluation, cross validation, model selection (hyperparameter search) - ensembles (from R) - model deployment (POJO export), fast scoring (<1ms)
34. install.packages("h2o") http://www.h2o.ai/
35. https://gist.github.com/szilard/b87233bbf41a4b366c26eede7bb1a0f3 Laptop / 1 server / cluster
39. No need for manual 1-hot encoding of categorical variables
57. https://gist.github.com/szilard/b87233bbf41a4b366c26eede7bb1a0f3