Purely Tensorflow, no Keras, no slim or other abstract libraries of Tensorflow. This repository focused on not-really deep architecture.
Some of notebooks got GIF showing training movement. WARNING, it pretty heavy.
Convolutional Neural Network
- Scratch Alex-net CIFAR 10
- Capsule Network
- Encoder-Decoder
- Residual Network
- Basic Conv on MNIST
- Byte-Net Translator
- Siamese Network on MNIST
- Generalized Hamming Network on MNIST
- Binary-net
- Kmax Conv1d
- Temporal Conv1d
- Triplet loss on MNIST
- Dense-net
Feed-forward Neural Network
- Batch-normalization
- Encoder-Decoder
- Word Vector
- Dropout Comparison, GIF included
- L1, L2, L1-L2 Regularization Comparison, GIF included
- Optimizer Comparison (Gradient Descent, Adagrad, RMSProp, Adam), GIF included
- Batch-normalization Comparison, GIF included
- Self-Normalized without and with API on MNIST
- Addsign and Powersign Optimizer
- Backprop without Learning Rates Through Coin Betting Optimizer (COCOB)
Recurrent Neural Network
- Music Generator
- Stock forecasting, GIF included
- Text Generator
- Signal Classifier
- Generator Comparison (LSTM GRU, LSTM Bidirectional, GRU Bidirectional), GIF included
- Time-Aware Long-Short Term Memory
- Dilated RNN
- Layer-Norm LSTM
- Neural Turing Machine
- Only Attention
- Multihead Attention
- Fast-slow LSTM
- Siamese Network
- Nested LSTM
- DNC (Differentiable Neural Computer)
- GAN Sentence
Attention Model
- Bahdanau
- Luong
- Hierarchical
- Additive
- Soft
- Attention-over-Attention
- Bahdanau API
- Luong API
Sequence-to-Sequence
- Basic Seq-to-Seq
- Beam decoder
- Chatbot with Attention (old API)
- Summarization with Attention (old API)
- Luong attention
- Bahdanau attention
- Bidirectional
- Estimator
- Altimatum (bidirectional + lstm + luong + beam)
Bayesian Hyperparameter Optimization
- Conv-CIFAR10
- Feedforward-Iris
- Recurrent-Sentiment
- Conv-Iceberg
Regression
- Linear Regression, GIF included
- Polynomial Regression, GIF included
- Ridge Regression, GIF included
- Lasso Regression, GIF included
- Elastic-net Regression, GIF included
- Sigmoid Regression, GIF included
- Quantile Regression
Reinforcement-learning
I code in external repository, can check here
- Policy gradient
- Q-learning
- Double Q-learning
- Recurrent-Q-learning
- Double Recurrent-Q-learning
- Dueling Q-learning
- Dueling Recurrent-Q-learning
- Double Dueling Q-learning
- Double Dueling Recurrent-Q-learning
- Actor-Critic
- Actor-Critic Dueling
- Actor-Critic Recurrent
- Actor-Critic Dueling Recurrent
- Async Q-learning
GAN
- DCGAN
- DiscoGAN
- Basic GAN
- WGAN-improve
Misc
- RNN-LSTM 20newsgroup Tensorboard histrogram
- Tensorboard debugger
- Transfer learning emotion dataset on MobilenetV2
- Multiprocessing tfrecords
- TF-Serving
tensorboard debugger
gradient techniques comparison
feed-forward, not dropout vs dropout