TensorFlow-LiveLessons
Note that the second edition of this video series is now available here. The second edition contains all of the content from this (first) edition plus quite a bit more, as well as updated library versions.
This repository is home to the code that accompanies Jon Krohn's:
- Deep Learning with TensorFlow LiveLessons (summary blog post here)
- Deep Learning for Natural Language Processing LiveLessons (summary blog post here)
- Deep Reinforcement Learning and GANs LiveLessons (summary blog post here)
The above order is the recommended sequence in which to undertake these LiveLessons. That said, Deep Learning with TensorFlow provides a sufficient theoretical and practical background for the other LiveLessons.
Prerequisites
Command Line
Working through these LiveLessons will be easiest if you are familiar with the Unix command line basics. A tutorial of these fundamentals can be found here.
Python for Data Analysis
In addition, if you're unfamiliar with using Python for data analysis (e.g., the pandas, scikit-learn, matplotlib packages), the data analyst path of DataQuest will quickly get you up to speed -- steps one (Introduction to Python) and two (Intermediate Python and Pandas) provide the bulk of the essentials.
Installation
Step-by-step guides for running the code in this repository can be found in the installation directory.
Notebooks
All of the code that I cover in the LiveLessons can be found in this directory as Jupyter notebooks.
Below is the lesson-by-lesson sequence in which I covered them:
Deep Learning with TensorFlow LiveLessons
Lesson One: Introduction to Deep Learning
1.1 Neural Networks and Deep Learning
- via analogy to their biological inspirations, this section introduces Artificial Neural Networks and how they developed to their predominantly deep architectures today
1.2 Running the Code in These LiveLessons
- goes over the installation directory mentioned above, discussing the options for working through my Jupyter notebooks
- details the step-by-step installation of TensorFlow on Mac OS X, a process that may be instructive for users of any Unix-like operating system
1.3 An Introductory Artificial Neural Network
- get your hands dirty with a simple-as-possible neural network (shallow_net_in_keras.ipynb) for classifying handwritten digits
- introduces Jupyter notebooks and their most useful hot keys
- introduces a gentle quantity of deep learning terminology by whiteboarding through:
- the MNIST digit data set
- the preprocessing of images for analysis with a neural network
- a shallow network architecture
Lesson Two: How Deep Learning Works
2.1 The Families of Deep Neural Nets and their Applications
- talk through the function and popular applications of the predominant modern families of deep neural nets:
- Dense / Fully-Connected
- Convolutional Networks (ConvNets)
- Recurrent Neural Networks (RNNs) / Long Short-Term Memory units (LSTMs)
- Reinforcement Learning
- Generative Adversarial Networks
2.2 Essential Theory I —- Neural Units
- the following essential deep learning concepts are explained with intuitive, graphical explanations:
- neural units and activation functions
- perceptron
- sigmoid (sigmoid_function.ipynb)
- tanh
- Rectified Linear Units (ReLU)
- neural units and activation functions
2.3 Essential Theory II -- Cost Functions, Gradient Descent, and Backpropagation
- cost functions
- quadratic
- cross-entropy (cross_entropy_cost.ipynb)
- gradient descent
- backpropagation via the chain rule
- layer types
- input
- dense / fully-connected
- softmax output (softmax_demo.ipynb)
2.4 TensorFlow Playground -- Visualizing a Deep Net in Action
- leverage TensorFlow Playground to interactively visualize the theory from the preceding section
2.5 Data Sets for Deep Learning
- overview of canonical data sets for image classification and meta-resources for data sets ideally suited to deep learning
2.6 Applying Deep Net Theory to Code I
- apply the theory learned throughout Lesson Two to create an intermediate-depth image classifier (intermediate_net_in_keras.ipynb)
- builds on, and greatly outperforms, the shallow architecture from Section 1.3
Lesson Three: Convolutional Networks
3.1 Essential Theory III -- Mini-Batches, Unstable Gradients, and Avoiding Overfitting
- add to our state-of-the-art deep learning toolkit by delving further into essential theory, specifically:
- weight initialization
- uniform
- normal
- Xavier Glorot
- stochastic gradient descent
- learning rate
- batch size
- second-order gradient learning
- momentum
- Adam
- unstable gradients
- vanishing
- exploding
- avoiding overfitting / model generalization
- L1/L2 regularization
- dropout
- artificial data set expansion
- batch normalization
- more layers
- max-pooling
- flatten
- weight initialization
3.2 Applying Deep Net Theory to Code II
- apply the theory learned in the previous section to create a deep, dense net for image classification (deep_net_in_keras.ipynb)
- builds on, and outperforms, the intermediate architecture from Section 2.5
3.3 Introduction to Convolutional Neural Networks for Visual Recognition
- whiteboard through an intuitive explanation of what convolutional layers are and how they're so effective
3.4 Classic ConvNet Architectures -— LeNet-5
- apply the theory learned in the previous section to create a deep convolutional net for image classification (lenet_in_keras.ipynb) that is inspired by the classic LeNet-5 neural network introduced in section 1.1
3.5 Classic ConvNet Architectures -— AlexNet and VGGNet
- classify color images of flowers with two very deep convolutional networks inspired by contemporary prize-winning model architectures: AlexNet (alexnet_in_keras.ipynb) and VGGNet (vggnet_in_keras.ipynb)
3.6 TensorBoard and the Interpretation of Model Outputs
- return to the networks from the previous section, adding code to output results to the TensorBoard deep learning results-visualization tool
- explore TensorBoard and explain how to interpret model results within it
Lesson Four: Introduction to TensorFlow
4.1 Comparison of the Leading Deep Learning Libraries
- discuss the relative strengths, weaknesses, and common applications of the leading deep learning libraries:
- Caffe
- Torch
- Theano
- TensorFlow
- and the high-level APIs TFLearn and Keras
- conclude that, for the broadest set of applications, TensorFlow is the best option
4.2 Introduction to TensorFlow
- introduce TensorFlow graphs and related terminology:
- ops
- tensors
- Variables
- placeholders
- feeds
- fetches
- build simple TensorFlow graphs (first_tensorflow_graphs.ipynb)
- build neurons in TensorFlow (first_tensorflow_neurons.ipynb)
4.3 Fitting Models in TensorFlow
- fit a simple line in TensorFlow:
- by considering individual data points (point_by_point_intro_to_tensorflow.ipynb)
- while taking advantage of tensors (tensor-fied_intro_to_tensorflow.ipynb)
- with batches sampled from millions of data points (intro_to_tensorflow_times_a_million.ipynb)
4.4 Dense Nets in TensorFlow
- create a dense neural net (intermediate_net_in_tensorflow.ipynb) in TensorFlow with an architecture identical to the intermediate one built in Keras in Section 2.5
4.5 Deep Convolutional Nets in TensorFlow
- create a deep convolutional neural net (lenet_in_tensorflow.ipynb) in TensorFlow with an architecture identical to the LeNet-inspired one built in Keras in Section 3.4
Lesson Five: Improving Deep Networks
5.1 Improving Performance and Tuning Hyperparameters
- detail systematic steps for improving the performance of deep neural nets, including by tuning hyperparameters
5.2 How to Built Your Own Deep Learning Project
- specific steps for designing and evaluating your own deep learning project
5.3 Resources for Self-Study
- topics worth investing time in to become an expert deployer of deep learning models
Deep Learning for Natural Language Processing
Lesson One: The Power and Elegance of Deep Learning for NLP
1.1 Introduction to Deep Learning for Natural Language Processing
- high-level overview of deep learning as it pertains to Natural Language Processing (NLP)
- influential examples of industrial applications of NLP
- timeline of contemporary breakthroughs that have brought Deep Learning approaches to the forefront of NLP research and development
1.2 Computational Representations of Natural Language Elements
- introduce the elements of natural language
- contrast how these elements are represented by traditional machine-learning models and emergent deep-learning models
1.3 NLP Applications
- specify common NLP applications and bucket them into three tiers of relative complexity
1.4 Installation, Including GPU Considerations
- build on the step-by-step installation of TensorFlow on Mac OS X covered in the Deep Learning with TensorFlow LiveLessons to facilitate the training of deep learning models with an Nvidia GPU.
1.5 Review of Prerequisite Deep Learning Theory
- summarise the key concepts introduced in the Deep Learning with TensorFlow LiveLessons, which serve as the foundation for the material introduced in these NLP-focused LiveLessons
1.6 A Sneak Peak
- take a tantalising look ahead at the capabilities developed over the course of these LiveLessons
Lesson Two: Word Vectors
2.1 Vector-Space Embedding
- leverage interactive demos to enable an intuitive understanding of vector-space embeddings of words, nuanced quantitative representations of word meaning
2.2 word2vec
- key papers that led to the development of word2vec, a technique for transforming natural language into vector representations
- essential word2vec theory introduced:
- architectures:
- Skip-Gram
- Continuous Bag of Words
- training algorithms:
- hierarchical softmax
- negative sampling
- evaluation perspectives:
- intrinsic
- extrinsic
- hyperparameters:
- number of dimensions
- context-word window size
- number of iterations
- size of data set
- architectures:
- contrast word2vec with its leading alternative, GloVe
2.3 Data Sets for NLP
- pre-trained word vectors:
- natural language data sets:
- Jon Krohn's resources page
- Zhang, Zhao and LeCun's labelled data
- Internet Movie DataBase (IMDB) reviews classified by sentiment from Andrew Maas and his Stanford colleagues (2011)
2.4 Creating Word Vectors with word2vec
- use books from Project Gutenberg to create word vectors with word2vec
- interactively visualise the word vectors with the bokeh library (creating_word_vectors_with_word2vec.ipynb)
Lesson Three: Modeling Natural Language Data
3.1 Best Practices for Preprocessing Natural Language Data
- in natural_language_preprocessing_best_practices.ipynb, apply the following recommended best practices to clean up a corpus natural language data prior to modeling:
- tokenize
- convert all characters to lowercase
- remove stopwords
- remove punctuation
- stem words
- handle bigram (and trigram) word collocations
3.2 The Area Under the ROC Curve
- detail the calculation and functionality of the area under the Receiver Operating Characteristic curve summary metric, which is used throughout the remainder of the LiveLessons for evaluating model performance
3.3 Dense Neural Network Classification
- pair vector-space embedding with the fundamentals of deep learning introduced in the Deep Learning with TensorFlow LiveLessons to create a dense neural network for classifying documents by their sentiment (dense_sentiment_classifier.ipynb)
3.4 Convolutional Neural Network Classification
- add convolutional layers to the deep learning architecture to improve the performance of the natural language classifying model (convolutional_sentiment_classifier.ipynb)
Lesson Four: Recurrent Neural Networks
4.1 Essential Theory of RNNs
- provide an intuitive understanding of Recurrent Neural Networks (RNNs), which permit backpropagation through time over sequential data, such as natural language and financial time series data
4.2 RNNs in Practice
- incorporate simple RNN layers into a model that classifies documents by their sentiment (rnn_in_keras.ipynb
4.3 Essential Theory of LSTMs and GRUs
- develop familiarity with the Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) varieties of RNNs which provide markedly more productive modeling of sequential data with deep learning models
4.4 LSTMs and GRUs in Practice
- straightforwardly build LSTM (vanilla_lstm_in_keras.ipynb) and GRU (gru_in_keras.ipynb) deep learning architectures through the Keras high-level API
Lesson Five: Advanced Models
5.1 Bi-Directional LSTMs
- Bi-directional LSTMs are an especially potent variant of the LSTM
- high-level theory on Bi-LSTMs before leveraging them in practice (bidirectional_lstm.ipynb)
5.2 Stacked LSTMs
- Bi-LSTMs are stacked to enable deep learning networks to model increasingly abstract representations of language (stacked_bidirectional_lstm.ipynb; ye_olde_conv_lstm_stackeroo.ipynb)
5.3 Parallel Network Architectures
- advanced data modeling capabilities are possible with non-sequential architectures, e.g., parallel convolutional layers, each with unique hyperparameters (multi_convnet_architectures.ipynb)
Deep Reinforcement Learning and GANs
Lesson One: The Foundations of Artificial Intelligence
1.1 The Contemporary State of AI
- examine what the term "Artificial Intelligence" means and how it relates to deep learning
- define narrow, general, and super intelligence
1.2 Applications of Generative Adversarial Networks
- uncover the rapidly-improving quality of Generative Adversarial Networks for creating compelling novel imagery in the style of humans
- involves the fun, interactive pix2pix tool
1.3 Applications of Deep Reinforcement Learning
- distinguish supervised and unsupervised learning from reinforcement learning
- provide an overview of the seminal contemporary deep reinforcement learning breakthroughs, including:
- the Deep Q-Learning algorithm
- AlphaGo
- AlphaGo Zero
- AlphaZero
- robotics advances
- introduce the most popular deep reinforcement learning environments:
1.4 Running the Code in these LiveLessons
- review the step-by-step installation of TensorFlow on Mac OS X detailed in the Deep Learning with TensorFlow LiveLessons
1.5 Review of Prerequisite Deep Learning Theory
- summarise the key concepts introduced in the Deep Learning with TensorFlow LiveLessons, which serve as the foundation for the material introduced in these advanced-topics LiveLessons
Lesson Two: Generative Adversarial Networks (GANs)
2.1 Essential GAN Theory
- cover the high-level theory of what GANs are and how they are able to generate realistic-looking images
2.2 The “Quick, Draw!” Game Dataset
- show the Quick, Draw! game, which we use as the source of hundreds of thousands of hand-drawn images from a single class for a GAN to learn to imitate
2.3 A Discriminator Network
- build the discriminator component of a GAN (generative_adversarial_network.ipynb)
2.4 A Generator Network
- build the generator component of a GAN (generative_adversarial_network.ipynb continued)
2.5 Training an Adversarial Network
- pit the generator and discriminator networks against each other as adversaries (generative_adversarial_network.ipynb completed)
Lesson Three: Deep Q-Learning Networks (DQNs)
3.1 The Cartpole Game
- introduce the Cartpole Game, an environment provided by OpenAI and used throughout the remainder these LiveLessons to train deep reinforcement learning algorithms
3.2 Essential Deep RL Theory
- delve into the essential theory of deep reinforcement learning in general
3.3 Essential DQN Theory
- delve into the essential theory of Deep Q-Learning networks, a popular, particular type of deep reinforcement learning algorithm
3.4 Defining a DQN Agent
- define a Deep Q-Learning agent from scratch (cartpole_dqn.ipynb)
3.5 Interacting with an OpenAI Gym Environment
- leverage OpenAI Gym to enable our Deep Q-Learning agent to master the Cartpole Game (cartpole_dqn.ipynb completed)
Lesson Four: OpenAI Lab
4.1 Visualizing Agent Performance
- use the OpenAI Lab to visualise our Deep Q-Learning agent's performance in real-time
4.2 Modifying Agent Hyperparameters
- learn to straightforwardly optimise a deep reinforcement learning agent's hyperparameters
4.3 Automated Hyperparameter Experimentation and Optimization
- automate the search through hyperparameters to optimize our agent’s performance
4.4 Fitness
- calculate summary metrics to gauge our agent’s overall fitness
Lesson Five: Advanced Deep Reinforcement Learning Agents
5.1 Policy Gradients and the REINFORCE Algorithm
- at a high level, discover Policy Gradient algorithms in general and the classic REINFORCE implementation in particular
5.2 The Actor-Critic Algorithm
- cover how Policy Gradients can be combined with Deep Q-Learning to facilitate the Actor-Critic algorithms
5.3 Software 2.0
- discuss how deep learning is ushering in a new era of software development driven by data in place of hard-coded rules
5.4 Approaching Artificial General Intelligence
- return to our discussion of Artificial Intelligence, specifically addressing the limitations of modern deep learning approaches