Pytorch Implementation Examples

This repository holds source codes of machine learning and modulized neural networks implemented with Pytorch.

Any comments or feedbacks are welcomed, email me at imhgchoi@korea.ac.kr

Contents

1. Gradient Descent : Not Pytorch — simple gradient descent with several different conditions.
2. Logistic Regression : Not Pytorch
3. Deep Neural Networks : predicting handwritten numbers with MNIST dataset
4. Convolutional Neural Networks : predicting handwritten numbers with MNIST dataset
5. Recurrent Neural Networks : predicting future stock price trend with RNN(LSTM cells)
6. AutoEncoders
   
   6.1 Feed Forward AutoEncoder : regenerating MNIST images with a feed forward AutoEncoder
   
   6.2 Convolutional AutoEncoder : regenerating MNIST images with a convolutional AutoEncoder
   
   6.3 Beta-Variational AutoEncoder : regenerating MNIST images with a Beta-Variational AutoEncoder
   
   I found it hard to build a vanilla VAE. So I adopted the Beta-VAE with an incremental Beta to help convergence.
   
   6.4 Sparse AutoEncoder : regenerating MNIST images with a sparse AutoEncoder with 1300 hidden code units.
   
   6.5 Denoising AutoEncoder : regenerating MNIST images that has gaussian noise with a denoising AutoEncoder.
7. Deep Q Network
   
   7.1 Feed Forward DQN : training Cartpole with an RL feed forward DQN
   
   7.2 Convolutional DQN : training Cartpole with an RL Convolutional DQN. Referenced here, but failed to master the game

   NOTE : All Neural Network Models are built without train/dev/test splits. Models will be prone to overfitting.