Agent and environment to apply RL in the bomberman game

Install procedure

Make sure that you are in the same folder as the setup.py. To achieve better results we recommend that you use a version of Python between 3.6 and 3.8, you might encounter errors if you run using newer versions.

pip install -e .

Demos

Run pre-trained agents

In this demo you'll run our pre-trained agents.

First, go to the scripts folder:

cd src/scripts

Then, run one of the agents:

python simulate_single_agent.py --agent-name dqn_agent
python simulate_single_agent.py --agent-name qagent
python simulate_single_agent.py --agent-name a2c_agent
python simulate_single_agent.py --agent-name policy_gradient_agent

Train agents from scratch

In this demo you'll train a Q-learning agent to destroy blocks.

First, go to the scripts folder:

cd src/scripts

Then, train one of the agents:

python train_dqn_agent.py --agent-name dqn_agent_new
python train_q_agent.py --agent-name qagent_new
python train_a2c_agent.py --agent-name a2c_agent_new
python train_policy_gradient_agent.py --agent-name policy_gradient_agent_new

It might be needed to train a bit more the agent, you can continue training with one of these commands:

python train_q_agent.py --agent-name qagent_new --agent-pretrained-name qagent_new
python train_dqn_agent.py --agent-name dqn_agent_new --agent-pretrained-name dqn_agent_new
python train_a2c_agent.py --agent-name a2c_agent_new --agent-pretrained-name a2c_agent_new
python train_policy_gradient_agent.py --agent-name policy_gradient_agent_new --agent-pretrained-name policy_gradient_agent_new

Run tests

python setup.py pytest