Deep-Learning-Tensorflow
Purely Tensorflow, no Keras or other abstract libraries of Tensorflow
Dependencies
sudo pip install scipy numpy matplotlib librosa pandas seaborn
- I recommended install Tensorflow from source, way more faster
- If you got GPU, compile it with CUDA
- You need to download CIFAR-10, CIFAR-100
Basic-Seq2Seq
Generate encoder and decoder by creating 2 Deep Recurrent Neural Network to predict incoming text link notebook
Chatbot-Attention-Seq2Seq
Generate chatbot using attention model on Sequence-to-Sequence Tensorflow API link notebook
DCGAN (Simplify and Original for House Number)
WGAN Improvement
DiscoGAN (original paper and Fashion MNIST)
Residual Network for CIFAR-10
Deep Convolutional
-
trained to label 100 classes link folder
-
trained to label multitags, a single picture can be more than 1 tag link notebook
-
trained to predict pokemon type link notebook
Deep Recurrent
-
trained to predict stock market link notebook
-
trained to generate sentence link notebook
-
trained to classify sentiment link notebook
Essay-Attention-Seq2Seq
Multi-Perceptron
-
Creditcard detection (softmax, l2 loss, 4 hidden layers) link notebook
-
detect-voice (softmax, dropout, l2 loss, 4 hidden layers) link notebook
-
iris (3 hidden layers, softmax) link notebook
-
pokemon (4 hidden layers, softmax) link notebook
-
sentiment (6 hidden layers, batch normalization, l2 loss, dropout) link notebook
Introduction on layer normalization
Encoder model, both multi-perceptron and Convolutional
-
multi-perceptron link notebook
-
Convolutional link notebook
Word vector both using softmax and NCE
-
softmax link notebook
-
NCE link notebook