rlib

rlib is a small deep reinforcement learning library with implementations of popular deep RL algorithms. Each algorithm is highly modular and customizable, making this library a great choice for anyone who wants to test the performance of different algorithms in the same environment. rlib uses PyTorch as the library of choice for its initial version, but support for TensorFlow is on the roadmap.

Installation

pip install rlib

Usage

Using rlib is this simple:

from rlib.algorithms.dqn import DQNAgent
from rlib.environments.gym import GymEnvironment


e = gym.make('CartPole-v0')

observation_size = 4
action_size = 2

dqn = DQNAgent(observation_size, action_size)
env = GymEnvironment(e, dqn)
env.train()
env.test()

Advanced

TensorBoard and GIFRecorder

1. Initialize Logger and/or GIFRecorder objects.

os.makedirs('your/log/dir', exist_ok=True)

logger = Logger(output_dir)
gifs_recorder = GIFRecorder(output_dir, duration=3.0)

2. Initialize a new environment using these objects.

env = GymEnvironment(e, dqn, logger=logger, gifs_recorder=gifs_recorder)

3. To check Tensorboard logs, run:

tensorboard --logdir=your/log/dir

Custom models

1. Define your own custom model.

class NeuralNet(torch.nn.Module):
    def __init__(self):
        super(NeuralNet, self).__init__()
        self.fc1 = nn.Linear(4, 8) 
        self.relu = nn.ReLU()
        self.fc2 = nn.Linear(8, 2)  

    def forward(self, x):
        out = self.fc1(x)
        out = self.relu(out)
        out = self.fc2(out)
        return out

2. Check the documentation for the algorithm you are using for the appropriate argument name. For DQN:

dqn = DQNAgent(
    observation_size, action_size,
    qnetwork_local=NeuralNet(),
    qnetwork_target=NeuralNet(),
)

Saving model weights

1. Set the model_output_dir argument when creating a new instance of an algorithm to the directory where you want your model to be saved.

Testing

To run all tests:

python -m unittest discover test/

Contributing

Installation

Feel free to open issues with any bugs found or any feature requests. Pull requests are always welcome for new functionality.

virtualenv -p python3 venv
cd rlib/
source venv/bin/activate
pip install -r requirements.txt

To make sure your installation worked, run one of the algorithms in the test folder:

python test/algorithms/dqn_test.py

License

rlib is released under the MIT License.