https://github.com/dllllb/neurips2021-nethack-raph

First place neural net based solution for the NeurIPS 2021 Nethack Challenge
https://github.com/dllllb/neurips2021-nethack-raph

nethack neural-networks reinforcement-learning

Last synced: about 1 year ago
JSON representation

First place neural net based solution for the NeurIPS 2021 Nethack Challenge

• Host: GitHub
• URL: https://github.com/dllllb/neurips2021-nethack-raph
• Owner: dllllb
• License: mit
• Created: 2022-04-01T18:20:18.000Z (about 4 years ago)
• Default Branch: master
• Last Pushed: 2022-06-20T13:59:26.000Z (almost 4 years ago)
• Last Synced: 2023-03-11T21:18:15.335Z (over 3 years ago)
• Topics: nethack, neural-networks, reinforcement-learning
• Language: Python
• Homepage:
• Size: 16.9 MB
• Stars: 2
• Watchers: 1
• Forks: 2
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          


  



This repository contains [RAPH](https://www.aicrowd.com/challenges/neurips-2021-the-nethack-challenge/teams/raph) team solution for the [NetHack Challenge](https://nethackchallenge.com/), organized on NeurIPS 2021 competition track. This solution won the **first place** among neural net based solutions.

## Setup

```

pipenv install

pipenv shell

```

## Run

run single evaluation with visual output

```

python3 playtest.py

```

evaluate 4096 episodes without visual output

```

python3 local_test.py

```

## Code structure

- out agent: ```agents/torchbeast_agent.py```  

- game state preprocessing, algorithms: ```nethack_raph/``` 

![Nethack Banner](img/nethack-challenge.jpeg)