https://github.com/paulkass/basic_pbt_demo

Implementation of the basic example of PBT in Rust from https://arxiv.org/abs/1711.09846
https://github.com/paulkass/basic_pbt_demo

machine-learning pbt population-based-training rust

Last synced: over 1 year ago
JSON representation

Implementation of the basic example of PBT in Rust from https://arxiv.org/abs/1711.09846

• Host: GitHub
• URL: https://github.com/paulkass/basic_pbt_demo
• Owner: paulkass
• Created: 2019-12-18T05:25:27.000Z (over 6 years ago)
• Default Branch: master
• Last Pushed: 2019-12-22T02:09:14.000Z (over 6 years ago)
• Last Synced: 2025-02-03T05:50:04.550Z (over 1 year ago)
• Topics: machine-learning, pbt, population-based-training, rust
• Language: Rust
• Homepage:
• Size: 24.4 KB
• Stars: 0
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Basic PBT Demo

Implementation of the basic example of PBT in Rust from https://arxiv.org/abs/1711.09846

## Rough Overview

The basic example from the article tries to optimize a function ![equation](https://latex.codecogs.com/gif.latex?1.2&space;-&space;(h_0&space;\theta_0^2+h_1&space;\theta_1^2)) where ![equation2](https://latex.codecogs.com/gif.latex?\vec{h}) represents the hyperparameters. 

The objective function that we are trying to optimize against is ![equation3](https://latex.codecogs.com/gif.latex?1.2&space;-&space;(\theta_0^2&space;+&space;\theta_1^2))

This code uses PBT as described in the paper with the `exploit` step being to choose the optimal `h` and `theta` from among the threads, and the `explore` step being modifying the parameters by adding a value drawn from normal distribution with mean `0` and standard deviation `0.1`.

## Run 

Run with by running `cargo run` in the main directory.