ML-algorithms in practice

Few examples of algorithms I've tried while learning ML. Hopefully it's easy to read and understand.

1. Policy gradient. First take at reinforcement learning. Basic policy gradient implementation using tensorflow and Breakout from openAI's gym. I tried to keep it as simple as possible with only minor exceptions.

2. Deep Q Network. Second take at reinforcement learning. Basic DQN implementation using tensorflow and Breakout from openAI's gym.

3. Predict next character with RNN. Tensorflow RNN (LSTM) biult to predict next character. After rnn is trained, it can generate new text based on some initial text input.

Things implemented simply

The idea is to make everything simple and readable. Take a concept and implement it from scratch (numpy in this case). If you manage to do that, perhaps that will help you to understand a thing or two about the concept.

1. Multivariate regression. Optimized using gradient decent with regularization (lambda).

2. Logistic regression. The same as before. Simple logistic regression.

3. Vanilla neural network. Multi-layer perceptron optimized using Adam, regularized using dropout.

4. Convolutional neural network (not finished, still working on it).

5. Recurrent neural network. Single hidden layer recurrent network used for MNIST clasification.

Gifs to look at

Generating text with LSTM	Playing Breakout