ghadiaravi13/RLStuff

A collection of reinforcement learning algorithm implementations

RLStuff

A collection of me playing around with Reinforcement Learning and other stuff. See more on my blog.

Implemented Algorithms

• Q Learning
• Deep Q Learning
• REINFORCE
• REINFORCE with Baseline
• Actor-Critic TD(0)
• Actor-Critic Forward-view TD(λ)
• Actor-Critic Backward-view TD(λ)
• Off-Policy Actor Critic
• Compatible Off-Policy Deterministic Actor-Critic with Q Critic
• Deep Deterministic Policy Gradient
• Proximal Policy Optimization

Interesting papers

• Deep Reinforcement Learning that Matters
• Implemmentation Matters in Deep Policy Gradients: A Case Study on PPO and TRPO
• What Matters In On-Policy Reinforcement Learning? A Large-Scale Empirical Study

My Blog Posts

Genetic Algorithms

• 8Queens_GA: Genetic Algorithm: 8 Queens Problem

Q Learning

• QLearning_DQN: A First Look at Reinforcement Learning
• Atari_DQN: Reinforcement Learning: Deep Q-Learning with Atari Games

Policy Gradients

• REINFORCE: Reinforcement Learning: An Introduction to Policy Gradients
• REINFORCE-Continuous: Policy Parameterization for a Continuous Action Space
• REINFORCE-Baseline: Policy Gradients: REINFORCE with Baseline
• Off-Policy_Policy_Gradient: Actor-Critic: Off-Policy Actor-Critic Algorithm

Actor-Critic

• Actor-Critic: Value Function Approximations
• Actor-Critic_TD_0: Actor-Critic: Implementing Actor-Critic Methods
• Actor-Critic_TD_Lambda_Forward: Actor-Critic: Implementing Actor-Critic Methods
• Actor-Critic_TD_Lambda_Backward: Actor-Critic: Implementing Actor-Critic Methods
• Off-Policy_Actor-Critic: Actor-Critic: Off-Policy Actor-Critic Algorithm

Deterministic Policy Gradients

• COPDAC-Q: Introduction to Deterministic Policy Gradient (DPG)

Policy Optimization Algorithms

• PPO_Discrete: Policy Optimizations: TRPO/PPO

Environments

ROMS:

• ROMs of Atari games I've used in my code. Note that with the latest version of OpenAI's gym, you need to import ROMs manually to run Atari environments.

ContinuousCartPole

• An implementation of CartPole with continuous action space by iandanforth