Entity Gym is an open source Python library that defines an entity based API for reinforcement learning environments. Entity Gym extends the standard paradigm of fixed-size observation spaces by allowing observations to contain dynamically-sized lists of entities. This enables a seamless and highly efficient interface with simulators, games, and other complex environments whose state can be naturally expressed as a collection of entities.
The enn-trainer library can be used to train agents for Entity Gym environments.
pip install entity-gym
You can find tutorials, guides, and an API reference on the Entity Gym documentation website.
A number of simple example environments can be found in entity_gym/examples. More complex examples can be found in the ENN-Zoo project, which contains Entity Gym bindings for Procgen, Griddly, MicroRTS, VizDoom, and CodeCraft.