PyCon India - Commodity Machine Learning; past, present and future

Transcript

1. 1.

   Commodity Machine Learning Past, present and future Andreas Mueller

2. 2.

   What is machine learning?

3. 3.

   Automatic Decision Making Spam? Yes No

4. 4.

   Spam? Yes No

5. 5.

   Programming Machine Learning

6. 6.

   Machine learning is EVERYWHERE

7. 7.

   None
8. 8.

   None
9. 9.

   None
10. 10.

    Science Engineering Medicine ...

11. 11.

    Commodity machine learning

12. 12.

    past

13. 13.

    +

14. 14.

    None
15. 15.

    dawn of open source tools...

16. 16.

    The age of shell

17. 17.

    Documentation? Testing?

18. 18.

    Scikit-learn: User centric machine learning

19. 19.

    .fit(X, y) .predict(X) .transform(X)

20. 20.

    present

21. 21.

    Choose your ecosystem.

22. 22.

    Open! Documented! Tested!

23. 23.

    Usability is key!

24. 24.

    ML Frameworks PyMC, Edward, Stan theano, tensorflow, keras

25. 25.

    None
26. 26.

    from sklearn.model_selection import GridSearchCV from sklearn.pipeline import Pipeline

27. 27.

    github.com/scikit­learn­contrib/scikit­learn­contrib

28. 28.

    (near) Future

29. 29.

    pip install scikit­learn==0.18rc2 0.18 for the release candidate:

30. 30.

    sklearn.cross_validation sklearn.grid_search sklearn.learning_curve sklearn.model_selection

31. 31.

    results = pd.DataFrame(grid_search.results_)

32. 32.

    labels → groups n_folds → n_splits

33. 33.

    from sklearn.cross_validation import KFold cv = KFold(n_samples, n_folds) for train,

    test in cv: ... from sklearn.model_selection import KFold cv = KFold(n_folds) for train, test in cv.split(X, y): ...
34. 34.

    from sklearn.mixture import GaussianMixture from sklearn.mixture import BayesianGaussianMixture

35. 35.

    PCA() RandomizedPCA() PCA()

36. 36.

    Gaussian Process Rewrite

37. 37.

    Isolation Forests

38. 38.

    Play from sklearn.neural_network import MLPClassifier Work import keras

39. 39.

    pipe = Pipeline([('preprocessing', StandardScaler()), ('classifier', SVC())]) param_grid = {'preprocessing': [StandardScaler(),

    None]} grid = GridSearchCV(pipe, param_grid)
40. 40.

    40

41. 41.

    (further) Future

42. 42.

    Feature / Column names

43. 43.

    from __future__ import sklearn.plotting

44. 44.

    from __future__ import AutoClassifier

45. 45.

    More Transparency

46. 46.

    amueller.github.io @amuellerml @amueller t3kcit@gmail.com