RL practice

This repository contains a set of algorithms and gyms for educational purposes and reinforcement learning experiments. The primary goal is to implement everything from scratch to gain a deeper understanding of RL.

Algorithms:

• DQN
• DoubleDQN
• DuelingDQN

GYMs:

• Snake

TODO:

• RainbowDQN implementation
• Platformer environment
• PPO (Proximal Policy Optimization) implementation

Install

Use python 3.10 and following packages:

$ pip install -r requirements.txt

HOWTO

Run script:

$ python run.py --param_file YOUR_PARAM_FILE --mode train --plot_results --display

For example

python run.py --param_file ./configs/double_dqn_snake.yaml --mode train --plot_results --display

Parameters_search:

python run.py --param_file ./configs/double_dqn_snake.yaml --mode parameters_search

Test:

python run.py --param_file ./configs/double_dqn_snake.yaml --mode test --display

Results

DQN Result

Double DQN Result

Dueling DQN Result