This repository contains an implementation of the model-free approaches in :
- Paper : Least-Squares Policy Iteration
- Authors : Lagoudakis and Parr
- Date : 2003
To install, clone this repository and execute the following commands :
$ cd rl-lspi
$ pip install -r requirements.txt
$ pip install -e .
The available policy evaluation methods are :
- LSTDQ (iterative or by batch)
- LSTDQ-OPT (iterative, based on Sherman-Morrison formula)
The available features are :
- Polynomial functions
- Radial Basis functions
The experiments in the paper are reproduced for the following environments :
- Experiment 1 : Chain Walk environment
- Experiment 2 : Inverted Pendulum environment