Regularization and Variance-Weighted Regression Achieves Minimax Optimality in Linear MDPs: Theory and Practice

This repository contains a collection of reproduction codes for the experiments described in Section 7 of the paper Regularization and Variance-Weighted Regression Achieves Minimax Optimality in Linear MDPs: Theory and Practice.

Requirements

First clone the repository with the following command.

git submodule update --init --recursive

Then move to the repository of each experiment and refer to its README.md.

Details of each experiment can be found in the following directories:

• VWLS-MDVI Experiment (Section 7.1): See VWLS-MDVI directory.
• Deep-Variance-Weighting in Maze (Section 7.2.1): See Deep-Variance-Weighting-Maze directory.
• Deep-Variance-Weighting in MinAtar (Section 7.2.2): See Deep-Variance-Weighting-MinAtar directory.