Getting Started with Machine Learning

Software is eating the world and ML is going to eat software. Many developers, both young and seasoned, want to get started with Machine Learning.

This is my deck for a talk I gave at Google Developer Group DevFest, 2018 at New Delhi. The main points covered are:

* ML Applications
* Neural Networks
* Tools
* Training your first classifier
* Resources
* Next Steps

Transcript

1. Getting Started with Machine Learning Avinash Hindupur @hindupuravinash
2. Agenda • ML Applications • Neural Networks • Tools • Train your first classifier • Resources • Next Steps
3. Applications • Image Classification • Neural style transfer on images and videos • Visual Question Answering • Image and Video Captioning • Text generation from a prompt • Story based question answering • Image generation - GANs • Games, deep RL
4. Neural Style Transfer Source; Yijun Liu et el
5. Visual QnA Source: Yash et el
6. Image captioning Source: Karpathy et el
7. Text Generation Source: Text-Generation-using-Bidirectional-LSTM-and-Doc2Vec-models
8. Story based QnA Source: Rajkumar et el
9. Image Generation Source: Karras et el
10. Reinforcement Learning Source: David Silver et el
11. A Simple Neural Network Source: James Loy
12. Variety of Neural Networks Source: Neural Network Zoo
13. Tools
14. Tensorflow TensorFlow is an open source software library for numerical computation using data flow graphs. • Nodes - mathematical operations • Edges - multidimensional data arrays (tensors)
15. Keras Keras is a high-level neural networks API. Written in Python and capable of running on top of TensorFlow, CNTK, or Theano. • Allows for easy and fast prototyping • Supports both CNNs and RNNs, as well as combinations of the two. • Runs seamlessly on CPU and GPU.
16. Training your first classifier • Load Data • Define Model • Compile Model • Fit Model • Evaluate Model • Predict
17. Load Data fashion_mnist = keras.datasets.fashion_mnist (train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data()
18. Loading Data Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-labels-idx1-ubyte.gz 32768/29515 [=================================] - 0s 0us/step Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-images-idx3-ubyte.gz 26427392/26421880 [==============================] - 1s 0us/step Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-labels-idx1-ubyte.gz 8192/5148 [===============================================] - 0s 0us/step Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-images-idx3-ubyte.gz 4423680/4422102 [==============================] - 0s 0us/step
19. Loading Data • Fashion-MNIST • 70,000 grayscale images • 10 categories • Each image is 28x28 Fashion-MNIST, Zalando
20. Explore Data train_images.shape (60000, 28, 28) len(train_labels) 60000 train_labels array([9, 0, 0, ..., 3, 0, 5], dtype=uint8) test_images.shape (10000, 28, 28) len(test_labels) 10000 'T-shirt/top' 'Trouser' 'Pullover' 'Dress' 'Coat' 'Sandal' 'Shirt' 'Sneaker' 'Bag' 'Ankle boot'
21. Preprocess Data plt.figure() plt.imshow(train_images[0]) plt.colorbar() plt.grid(False) train_images = train_images / 255.0 test_images = test_images / 255.0
22. Preprocess Data plt.figure(figsize=(10,10)) for i in range(25): plt.subplot(5,5,i+1) plt.xticks([]) plt.yticks([]) plt.grid(False) plt.imshow(train_images[i], cmap=plt.cm.binary) plt.xlabel(class_names[train_labels[i]])
23. Define Model model = keras.Sequential([ keras.layers.Flatten(input_shape=(28, 28)), keras.layers.Dense(128, activation=tf.nn.relu), keras.layers.Dense(10, activation=tf.nn.softmax) ])
24. Compile Model model.compile(optimizer=tf.train.AdamOptimizer(), loss='sparse_categorical_crossentropy', metrics=['accuracy'])
25. Fit Model model.fit(train_images, train_labels, epochs=5)
26. Fit Model Epoch 1/5 60000/60000 [==============================] - 5s 90us/step - loss: 0.5001 - acc: 0.8253 Epoch 2/5 60000/60000 [==============================] - 5s 88us/step - loss: 0.3761 - acc: 0.8644 Epoch 3/5 60000/60000 [==============================] - 5s 89us/step - loss: 0.3384 - acc: 0.8759 Epoch 4/5 60000/60000 [==============================] - 5s 89us/step - loss: 0.3124 - acc: 0.8857 Epoch 5/5 60000/60000 [==============================] - 5s 86us/step - loss: 0.2946 - acc: 0.8922
27. Evaluate Model test_loss, test_acc = model.evaluate(test_images, test_labels) print('Test accuracy:', test_acc) Test accuracy: 0.8777
28. Predict predictions = model.predict(test_images) predictions[0] array([4.0652740e-06, 6.9819279e-08, 2.5388722e-06, 1.3390627e-07, 1.1847248e-07, 2.9022932e-02, 2.0918555e-06, 6.4492501e-02, 9.1468155e-06, 9.0646631e-01], dtype=float32) np.argmax(predictions[0]) 9
29. Predict i = 0 plt.figure(figsize=(6,3)) plt.subplot(1,2,1) plot_image(i, predictions, test_labels, test_images) plt.subplot(1,2,2) plot_value_array(i, predictions, test_labels)
30. Predict # Plot the first X test images, their predicted label, and the true label # Color correct predictions in blue, incorrect predictions in red num_rows = 5 num_cols = 3 num_images = num_rows*num_cols plt.figure(figsize=(2*2*num_cols, 2*num_rows)) for i in range(num_images): plt.subplot(num_rows, 2*num_cols, 2*i+1) plot_image(i, predictions, test_labels, test_images) plt.subplot(num_rows, 2*num_cols, 2*i+2) plot_value_array(i, predictions, test_labels)
31. Predict
32. Resources • Deep Learning Book - A free practical book on deep learning by Ian Goodfellow • Fast AI course - ML crash course for coders • Deep Learning AI - Learn deep learning from Andrew Ng • Deep Hunt - A newsletter on latest developments in ML • Tensorflow Tutorials - Official guides
33. What can you do next? **
34. What can you do next? • Keep practising - Kaggle, Open Source • Do online courses • Read blogs • Follow important researchers • Be active in community
35. Thank you!