/Sparse-IL

Code accompanying the paper: Action Assembly: Sparse Imitation Learning for Text Based Games with Combinatorial Action Spaces

Primary LanguageJupyter NotebookGNU General Public License v3.0GPL-3.0

Sparse-IL

Code accompanying the paper: Action Assembly: Sparse Imitation Learning for Text Based Games with Combinatorial Action Spaces

ZorkGym contains the Gym environment for the Zork domain.

deep_compressed_sensing.py - Train a DeepCS agent. imitation_learning_all.py - Train the Sparse-IL agent.

Notebooks directory contains the various notebooks used to evaluate the trained agents.