Rainbow

Implementation of Rainbow paper sans distributional RL as part of Reinforcement Learning course at Chula 2019

This repository contains the solution notebooks of all Rainbow improvements except distributional reinforcement learning.

The solutions folder contains

• environments.py - The SingleStockEnvironment for single-asset trading (buy, hold, sell)
• agents.py - Agents for function-approximation Q-learning and deep Q-learning
• networks.py - Networks used by the agents
• memories.py - Memories used by the agents
• utils.py - Some utility functions

The notebooks are

• qlearning_fa.ipynb - Function-approximation Q-learning
• dqn_vanilla.ipynb - Vanilla deep Q-learning
• dqn_double.ipynb - Double deep Q-learning
• dqn_prioritized.ipynb - DQN with prioritized memory
• dqn_nstep.ipynb - DQN with N-step memory
• dqn_dueling.ipynb - DQN with dueling networks
• dqn_noisy.ipynb - DQN with noisy linear layers for exploration
• dqn_rainbow.ipynb - Rainbow implementation without distributional RL
• dqn_trading.ipynb - Use Rainbow to trade bitcoins