https://github.com/instance01/grab0
It GRABs you, it consumes you. What is GRAB doing you ask? It is learning. With 0 domain knowledge.
https://github.com/instance01/grab0
Last synced: 3 months ago
JSON representation
It GRABs you, it consumes you. What is GRAB doing you ask? It is learning. With 0 domain knowledge.
- Host: GitHub
- URL: https://github.com/instance01/grab0
- Owner: instance01
- Created: 2020-07-20T21:06:20.000Z (over 5 years ago)
- Default Branch: master
- Last Pushed: 2020-12-08T13:38:06.000Z (about 5 years ago)
- Last Synced: 2025-01-20T00:34:22.268Z (about 1 year ago)
- Language: C++
- Homepage:
- Size: 21.1 MB
- Stars: 3
- Watchers: 2
- Forks: 0
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
Based on [bootleg AlphaZero](https://github.com/instance01/BootlegAlphaZero).
This repository implements GRAB0 with two approaches - GBTS (Gradient Bandit
Tree Search), which is MCTS but with Gradient Bandits, and SGBS (Stacked
Gradient Bandit Search) with a stack of Gradient Bandits.
Below is SGBS in a nutshell:
Experimental results for GRAB0 can be found in
[results.md](https://github.com/instance01/GRAB0/blob/master/results.md). There
are a few more that are currently not pushed yet, more information will follow
here.
Below is the result of training 33 runs on Cartpole using the GRAB0 approach in
combination with SGBS.
### Installation
Run `./setup.sh`. The `src` folder will contain a new binary `GRAB0`. You can adjust experiments in `simulations.json`.