Python modules allowing to train DQN agent, observe agent's games and collect its stats.

Report: https://www.overleaf.com/read/khmgdtdxtmhk

main.py

Train convolutional selected model on environment.
Currently doesn't support arguments.

test_agent_play.py

Supported arguments:

• -a, --agent
  Supported agents:
  • tDQN - Deep Q Learning in Pytorch
  • tAC - Actor Critic in Pytorch
  • kDQN - Deep Q Learning in Keras
• --model [PATH] - path to saved model's weights
  Agents currently supporting default weights:
  • tAC

Watch trained agent play snake or collect stats about agent's performance

Installation

This project requires snake_gym repo to be present in the same directory as this one.
https://github.com/Gizzio/snake_gym
The tree should look like this:

• your_dir
  • snake_gym
  • snake-rl

For example:

mkdir snake
cd snake
git clone git@github.com:DoomCoder/snake-rl.git
git clone git@github.com:Gizzio/snake_gym.git

1. Setup directories as descibed above
2. Install requirements.txt from snake_gym
3. Install requirementx.txt from snake-rl
   If your machine doesn't support CUDA, use requirements_cpu.txt
4. Done!