Pytorch implementation of the Influence-aware Memory Architectures for Deep Reinforcement Learning by Suau et al using pytorch-a2c-ppo-acktr-gail by Ilya Kostrikov.
To use please run the following command to install the required packages:
pip install -r requirements.txt
To be able to run the warehouse environment it needs to be registered to gym for new processes. This can be done by adding the following line to the gym.envs.__init__.py file:
register("Warehouse-v0", entry_point="environments.warehouse.warehouse:Warehouse", kwargs={"seed": 0, "parameters": {"num_frames": 1}})
The Warehouse envorinment can be run with a parallel IAM architecture, GRU architecture, just linear layers, or linear layers with 8 frames stacked using, respectively, the commandline arguments IAM, RNN, FNN or FNN8 on the file main.py.
A Google Colab notebook that sets up the environment as needed, trains on the Warehouse environment and saves the results to Google Drive can be found here: https://colab.research.google.com/drive/1nlj8GNdoFGpXgeY29Q7ZfjNc7A_0PEyL?usp=sharing
All results of the reproduction including plotting and raw data can be found in the results folder.
Preview of the warehouse environemnt:
