This is a currently ongoing project where I am building the building blocks of deep reinforcement learning from scratch, starting with the Neural Networks. This repository currently has:

• An implementation of Neural Networks
• An implementation of Backpropagation
• Losses:
  • Mean Squared Error
• Activation Functions:
  • ReLu
  • Sigmoid
• An implementation of DQN
• An implementation of Vanilla REINFORCE (Still Buggy)
• An implementation of a basic Actor-Critica algorithm (Still Buggy)

This is currently a work in progress. Not all these algorithms are working exactly as intended yet. However, since Reinforcement Learning algorithms tend to be quite inconsistent, especially with small model sizes, sometimes it pays to try again and see if the result is better.