Machine Learning without PhD :)

Байрам Аннаков @ MoscowPython Meetup 49

"Расскажем и покажем, как обучать машины и нейросети для решения простых и не очень задач классификации и распознавания".

Видео: http://www.moscowpython.ru/meetup/49/ml-no-phd/

Transcript

1. Machine Learning
   without PHD
   App in the Air
   

   View Slide

2. It all started
   with…
   

   View Slide

3. How?
   Reproduce the same in Python
   Train your own model
   

   View Slide

4. Reproduce
   

   View Slide

5. ImageNet
   

   View Slide

6. View Slide

7. Recognize image
   from keras.applications.resnet50 import ResNet50
   from keras.preprocessing import image
   from keras.applications.resnet50 import
   preprocess_input, decode_predictions
   import numpy as np
   model = ResNet50(weights='imagenet')
   

   View Slide

8. Recognize image
   img_path = 'elephant.jpg'
   img = image.load_img(img_path,
   target_size=(224, 224))
   x = image.img_to_array(img)
   x = np.expand_dims(x, axis=0)
   x = preprocess_input(x)
   

   View Slide

9. Recognize image
   preds = model.predict(x)
   # decode the results into a list of tuples (class,
   description, probability)
   # (one such list for each sample in the batch)
   print('Predicted:',
   decode_predictions(preds, top=3)[0])
   # Predicted: [(u'n02504013', u'Indian_elephant',
   0.82658225), (u'n01871265', u'tusker', 0.1122357),
   (u'n02504458', u'African_elephant', 0.061040461)]
   

   View Slide

10. CNN
    

    View Slide

11. View Slide

12. Train your own
    

    View Slide

13. https:/
    /blog.keras.io/
    

    View Slide

14. 72%
    

    View Slide

15. VGG16
    

    View Slide

16. +2 months —> 90%
    GPU!!
    Re-training more layers
    Different architecture
    Data augmentation
    Dropout
    

    View Slide

17. 2 internal seminars
    

    View Slide

18. Passport Scanner
    87%
    Rhythmic
    Gymnastics
    98,9%
    

    View Slide

19. Rhythmic Gymnastics
    

    View Slide

20. Approach
    • Collect data

    Simple iPhone app that helps draw and export
    • Prepare data

    Image = Grid. Each cell = 1 (black) or 0 (white)

    Convert Grid to Line

    Image = 000100011000011100011…
    • Train + Analyze

    Until satisfied with the score
    

    View Slide

21. View Slide

22. Prepare data
    import skimage

    import numpy
    

    View Slide

23. Train and Analyze
    1. K-neighbors 78%
    from sklearn.neighbors import KNeighborsClassifier
    clf = KNeighborsClassifier(n_neighbors=1)
    clf.fit(x_train, y_train)
    clf.score(x_test, y_test)
    

    View Slide

24. Train and Analyze
    2. K-neighbors + PCA 81%
    from sklearn.decomposition import PCA
    pca = PCA(n_components=40, whiten=True)
    pca.fit(x_train)
    x_train_pca = pca.transform(x_train)
    x_test_pca = pca.transform(x_test)
    //repeat KNN
    

    View Slide

25. May be someone has
    already solved it?
    

    View Slide

26. MNIST
    http://yann.lecun.com/exdb/mnist/
    

    View Slide

27. 3. SVM 90%
    from sklearn import svm
    classifier = svm.SVC(gamma=0.001)
    classifier.fit(x_train, y_train)
    predicted = classifier.predict(x_test)
    

    View Slide

28. Passport OCR
    Data
    char74
    icdar2003
    Architecture
    Modified MNIST
    

    View Slide

29. http:/
    /course.fast.ai
    

    View Slide

30. python@appintheair.mobi
    

    View Slide