plz717/reinforcement_learning

Implementations of selected reinforcement learning algorithms in Tensorflow. A3C, DDPG, REINFORCE, DQN, etc.

Implementations of Reinforcement Learning Algorithms in Python

Implementations of selected reinforcement learning algorithms with tensorflow and openai gym. Working examples.

Implemented Algorithms

(Click into the links for more details)

Advanced

• Asynchronized Advantage Actor-Critic (A3C)
• Deep Deterministic Policy Gradient (DDPG)

Policy Gradient Methods

• REINFORCE with policy function approximation
• REINFORCE with baseline

Temporal Difference Learning

• Standard epsilon greedy Q-learning
• Deep Q-learning

Monte Carlo Methods

• Monte Carlo (MC) estimation of action values

Dynamic Programming MDP Solver

• Value iteration
• Policy iteration - policy evaluation & policy improvement

OpenAI Gym Examples

• Cartpole-v0
  • TD/cartpole_qlearning.py: solved cartpole-v0 after 1598 episodes of training
  • DQN/cartpole_dqn.py: solved cartpole-v0 after 75 episodes of training
  • policy_gradient/cartpole_policy_gradient.py: REINFORCE solved cartpole-v0 after 632 episodes

• Pendulum-v0
  • ddpg/pendulum_ddpg.py: Best 100-episode average reward was -148.19 ± 9.14

Environments

• envs/gridworld.py: minimium gridworld implementation for testings

Dependencies

• Python 2.7
• Numpy
• Tensorflow 0.12.1
• OpenAI Gym (with Atari) 0.8.0
• matplotlib (optional)

Tests

• Files: test_*.py
• Run unit test for [class]:

python test_[class].py

MIT License