https://github.com/imoneoi/xrl-script

Efficient AutoRL script for any framework
https://github.com/imoneoi/xrl-script

auto-ml deep-reinforcement-learning distributed hyperparameter-optimization hyperparameter-search parallel reinforcement-learning

Last synced: 7 months ago
JSON representation

Efficient AutoRL script for any framework

• Host: GitHub
• URL: https://github.com/imoneoi/xrl-script
• Owner: imoneoi
• License: mit
• Created: 2020-08-05T02:49:13.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2020-08-05T03:16:14.000Z (over 5 years ago)
• Last Synced: 2024-10-29T12:37:47.475Z (over 1 year ago)
• Topics: auto-ml, deep-reinforcement-learning, distributed, hyperparameter-optimization, hyperparameter-search, parallel, reinforcement-learning
• Language: Python
• Homepage:
• Size: 5.86 KB
• Stars: 5
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# XRL-Script

Efficient distributed AutoRL script for any framework

## Features

- [x] Distributed training on single machine, multiple CPU/GPUs

- [x] Automatic resource allocation

- [x] TPESampler and Hyperband pruner support

- [ ] Automatic testing for stable hyperparameter set

## Usage

*In rl.py*

1. Implement your training logic in train_rl_agent

2. Specify hyperparameter range

*In rl-auto-gpu.py*

1. Specify minimum and maximum steps per trial, and number of trials

2. Set parallel trials per GPU & parallel envs per trial according to your hardware

3. Set reduction factor (integer). In most situations you can set one so that Hyperband bracket number stays in [4, 6]

**Then run rl-auto-gpu.py, the script will find all available gpus and run hyperparameter search in parallel**

## Dependencies

- [Optuna](https://github.com/optuna/optuna)