filangelos / social_rl

Social Reinforcement Learning

Installation

pip install -e .[dev]

Notes

This repository is under heavy refactoring, please do not raise issues unless you catch any breaking bugs. The list of changes that are under consideration are the following:

• Remove any dm-reverb dependencies.
• Host demonstrations on an Oxford server for guaranteeing reproducibility.
• Un-.gitignore the experiments/run_psiphi.ipynb after testing sensitivity to the l1_loss_coef parameter.
• Add stop_successor_features_gradients parameter to the ITDLoss.
• Consider trading scalability with readability for the loss_outputs in the _loss_fns.
• Remove any markov dependencies.