Home

Awesome

PWC ArXiv Badge

Rational Reinforcement Learning Repository

Using Rational Networks in Simple Reinforcement Learning (Atati task so far). This repository is used in the Recurrent Rational Networks publication.

rl scores

Dependencies

This Repository depends on:

Installation

First, please clone this repo and go into it:

git clone https://github.com/ml-research/rational_rl
cd rational_rl

A Dockerfile is provided, to create a docker image, please run:

docker build -t rationalrl . # to create a docker image
docker run -ti --gpus all -v $(pwd):/home/rl_paus rationalrl bash

This last command will instantiate a container from your image and run bash into it.*

*You need to have nvidia-docker installed to run docker containers with GPU and CUDA support (otherwise, please drop --gpu all).

Watch a trained agent play:

To watch a trained Recurrent Rational agent on Kangaroo, please provide its path: <br/> python3 rendering_atari.py updated_agents/DQN_recrat_Kangaroo_s0_e500.zip Hereafter are provided some compiled example of DQN Agents (left with Leaky ReLU, center with Rational and right with Recurrent Rationals)

you can find more gifs in videos/gifs_files/optim/Asterix

Usage

To get the trained agents, please contact Quentin Delfosse