This repository contains a small example of how to use an OpenAI Gym environment with EntityGym/ENN-Trainer.
- minefield.py file contains a OpenAI Gym version of the EntityGym Minefield example environment.
- gym_env_adapter.py implements a generic adapter for OpenAI Gym environments to EntityGym environments.
- train.py is a script that uses trains an agent on either the OpenAI Gym or EntityGym version of the Minefield environment with enn-trainer.
To install dependencies with poetry:
poetry install
poetry run pip install torch==1.12.0+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html
poetry run pip install torch-scatter -f https://data.pyg.org/whl/torch-1.12.0+cu113.htmlRunning training with the OpenAI Gym environment implementation:
poetry run python minefield/train.py env.id=openai_gym.Minefield net.translation='(reference_entity: "Vehicle", position_features: ["_0", "_1"], rotation_angle_feature: "_2")'Running training with the EntityGym implementation:
poetry run python minefield/train.py env.id=entity_gym.Minefield net.translation='(reference_entity: "Vehicle", position_features: ["x_pos", "y_pos"], rotation_angle_feature: "direction")'Comparison in W&B: https://wandb.ai/entity-neural-network/enn-ppo/reports/OpenAI-Gym-vs-EntityGym-Minefield-implementation--VmlldzoyNTQwMDI0