Competitive Training Methods

Competitive training methods for reinforcement learning using RLLIB

Requirements

Optionally create a Python3 virtualenv called venv (separate project dependencies)
```
 virtualenv -p python3 venv
```
Activate the virtualenv (if you created one)
```
 source venv/bin/activate
```
Install dependencies
```
 pip install -r requirements.txt
```

Activate the virtualenv (if used)

source venv/bin/activate

Run using main.py and configure via optional command line arguments

python main.py {options}

Flag	Parameters	Description	Required	Default Value
run	str	An optional name of this experiment run	N	''
epochs	int	The number of experiment epochs to run	N	1
env	str	The game environment to use	N	Connect4
agent	str	The agent types to use	N	Deep Q-Network (DQN)
trainer	str	The agent training method to use	N	DefaultTrainer
save-agent	bool	Whether to save all agents weights/memory/progress.	N	False
human-player	bool	Allows you to compete against trained agents.	N	False
visualise	bool	Whether to show a visualisation of the game.	N	False

Name		Name	Last commit message	Last commit date
Latest commit History 152 Commits
src		src
.gitignore		.gitignore
README.md		README.md
assumption_based_learning.py		assumption_based_learning.py
bandit_trainer.py		bandit_trainer.py
competitive_training.py		competitive_training.py
elo_evaluator.py		elo_evaluator.py
main.py		main.py
mcts_evaluator.py		mcts_evaluator.py
mcts_evolution.py		mcts_evolution.py
mcts_trainer.py		mcts_trainer.py
modified_episode.py		modified_episode.py
original_policy_mapping_fn.py		original_policy_mapping_fn.py
plateau-agentpool.py		plateau-agentpool.py
plateau-selfplay.py		plateau-selfplay.py
requirements.txt		requirements.txt
trainer_evaluator.py		trainer_evaluator.py