Interact contains implementations of several deep reinforcement learning algorithms.
Interact can be installed as follows:
git clone https://github.com/rystrauss/interact
cd interact
pip install .If you want to use Gym environments that aren't installed by default with Gym, you'll
need to install those yourself (e.g. pip install gym[atari]).
An agent can be trained with the following command:
python -m interact.train --config <path_to_config_file>
This package uses Gin to configure experiments,
and the --config option should be a path to a Gin config file. Algorithm-specific
arguments can be found in each agent's documentation.
Some example configuration files can be found in the examples directory.
Once an agent has been trained, it can be visualized in its environment with the following command:
python -m interact.play --agent_dir <path/to/agent/dir>
where <path/to/agent/dir> is the path to the directory that contains the agent you
want to visualize (this is the directory that was created by the training script).