/SORB

Experiment friendly non-ipython notebook version of "Search on the Replay Buffer"

Primary LanguagePythonMIT LicenseMIT

Search on the Replay Buffer

Experiment friendly non-iPython notebook version of the paper "Search on the Replay Buffer : Bridging Planning and Reinforcement Learning" by Eysenbach et al.

Installation

Install TF-Agents on top of the standard TF suite.

pip install --user --upgrade tf-agents-nightly

Code Structure

Majority of the code corresponding to the buffer and graph creation is in main.py. Comments with @param are hyperparameters.