Introduction to deep learning using python
•
2 likes•155 views
A brief introduction to machine learning and a short example on how to use Python and TensorFlow to build a neural network.
More Related Content
What's hot
What's hot (13)
Similar to Introduction to deep learning using python
Similar to Introduction to deep learning using python (20)
Recently uploaded
Recently uploaded (20)
Introduction to deep learning using python
- 1. Introduction to Deep Learning using Python Lino Coria lino@wiivv.com @linocoria
- 2. About Wiivv Wiivv is a technology company transforming footwear and apparel for every human body. Wiivv Insoles and Sandals are created uniquely for you, based on measurements taken from the award-winning Wiivv app.
- 3. Testing our sandals - Boston Marathon 2018
- 4. What is Deep Learning?
- 5. AI, Machine Learning, Deep Learning Artificial Intelligence Machine Learning Deep Learning
- 6. Artificial Intelligence - Deep Learning
- 8. The Power of Deep Neural Networks Amount of Data Performance Traditional ML Small NN Medium NN Large NN
- 9. Why is Deep Learning in Vogue? ● Hardware ○ GPUs ○ NVIDIA leading the way ● Tons of Data ○ ImageNet dataset: 1.4 million annotated images ● Better Algorithms ● Democratic ○ If you know Python, you can do deep learning ○ Many tutorials, pre-trained models
- 10. Deep Learning and Python
- 11. Open-Source Resources for Deep Learning Keras
- 12. A good way to start Keras
- 13. How to get Data
- 14. Example: MNIST Dataset A classic Machine Learning problem Input X (Images) Labels Y: 5 0 4 (Answers)
- 15. Training Data + Testing Data Input X (Images) Labels Y Training Data Input X (Images) Labels Y Testing Data
- 16. Neural Network 1. Input (image pixels) 2. Hidden layers (data transformations) 3. Output (predicted label)
- 17. Weights Layer (data transformation) Input X Weights Layer (data transformation) Predictions Y’ True targets Y (Labels) Loss function Loss score Optimizer Weight update
- 18. Load MNIST Data >>> import keras >>> from keras.datasets import mnist >>> (train_images, train_labels), (test_images, test_labels) = mnist.load_data() >>> train_images.shape (60000, 28, 28) >>> len(train_labels) 60000 Training input: 60,000 images (28 pixels * 28 pixels) Training labels: 60,000
- 19. MNIST training data >>> import matplotlib.pyplot as plt >>> plt.imshow(train_images[0]) >>> plt.gray() >>> plt.show() >>> train_labels array([5, 0, 4, ..., 5, 6, 8], dtype=uint8)
- 20. MNIST testing data >>> test_images.shape (10000, 28, 28) >>> test_labels array([7, 2, 1, ..., 4, 5, 6], dtype=uint8) Testing input: 10,000 images (28 pixels * 28 pixels) Testing labels: 10,000 Training 60,000 images 60,000 labels Testing 10,000 images 10,000 labels
- 21. Data Reshaping (Input X) # Train images >>> train_images = train_images.reshape((60000, 28 * 28)) >>> train_images = train_images.astype('float32') / 255 # Test images >>> test_images = test_images.reshape((10000, 28 * 28)) >>> test_images = test_images.astype('float32') / 255 1. 28 * 28 image (int values 0 - 255) 2. 28 * 28 = 784-element vector: [row 0, row 1, …, row 27] 3. 784-element vectors with float values in range 0 to 1
- 22. Data Reshaping (Labels Y) >>> from keras.utils import to_categorical # Train labels >>> train_labels = to_categorical(train_labels) # Test labels >>> test_labels = to_categorical(test_labels) 1. train_labels = [5 0 4 ... 5 6 8] 2. train_labels = [[0. 0. 0. 0. 0. 1. 0. 0. 0. 0.] [1. 0. 0. 0. 0. 0. 0. 0. 0. 0.] [0. 0. 0. 0. 1. 0. 0. 0. 0. 0.] … [0. 0. 0. 0. 0. 1. 0. 0. 0. 0.] [0. 0. 0. 0. 0. 0. 1. 0. 0. 0.] [0. 0. 0. 0. 0. 0. 0. 0. 1. 0.]]
- 23. Building our Network >>> from keras import models >>> from keras import layers >>> network = models.Sequential() >>> network.add(layers.Dense(512, activation='relu', input_shape=(28 * 28,))) >>> network.add(layers.Dense(512, activation='relu')) >>> network.add(layers.Dense(10, activation='softmax'))
- 24. Neural Network 1. Input (28 * 28 = 784) 2. Hidden layers (data transformations, 512 elements per layer) 3. Output (predicted label: 10-element vector with probabilities)
- 25. Weights Layer (data transformation) Input X Weights Layer (data transformation) Predictions Y’ True targets Y (Labels) Loss function Loss score Optimizer Weight update
- 26. Network compilation >>> network.compile(optimizer='rmsprop', ... loss='categorical_crossentropy', ... metrics=['accuracy'])
- 27. Weights Layer (data transformation) Input X Weights Layer (data transformation) Predictions Y’ True targets Y (Labels) Loss function Loss score Optimizer Weight update
- 28. Let’s train! >>> network.fit(train_images, train_labels, epochs=5, batch_size=128) Epoch 1/5 60000/60000 [==============================] - 2s 33us/step - loss: 0.2179 - acc: 0.9329 Epoch 2/5 60000/60000 [==============================] - 2s 31us/step - loss: 0.0805 - acc: 0.9750 Epoch 3/5 60000/60000 [==============================] - 2s 32us/step - loss: 0.0531 - acc: 0.9841 Epoch 4/5 60000/60000 [==============================] - 2s 31us/step - loss: 0.0385 - acc: 0.9883 Epoch 5/5 60000/60000 [==============================] - 2s 31us/step - loss: 0.0290 - acc: 0.9908 Train Accuracy: 99.08%
- 29. Test Accuracy >>> test_loss, test_acc = network.evaluate(test_images, test_labels) 10000/10000 [==============================] - 0s 47us/step >>> print('Test accuracy: %s' % test_acc) Test accuracy: 0.9799 Test Accuracy: 97.99%
- 30. Resources Online Udacity, Coursera PyImageSearch Book Deep Learning with Python by Francois Chollet