Skip to content

Latest commit

 

History

History
29 lines (25 loc) · 1.44 KB

File metadata and controls

29 lines (25 loc) · 1.44 KB
title JAX Implementation of Hindsight Experience Replay (HER)
categories
Research
Software
Tutorial
tags
machine-learning
programming
github jeertmans/HER-with-JAX
website https://colab.research.google.com/github/jeertmans/HER-with-JAX/blob/main/bit_flipping.ipynb
image
path alt
/assets/img/posts/her-with-jax.jpg
HER paper cover with JAX and Jupyter logos
permalink /posts/her-with-jax/
description Implementation of the Hindsight Experience Replay (HER) method in JAX.

I recently discovered the Hindsight Experience Replay (HER) paper and noticed that the official implementation is based on PyTorch and is not very well-structured. I also couldn't find a non-PyTorch implementation. Since I primarily work with JAX, I decided to reimplement the classic bit-flipping experiment to better understand HER.

This implementation uses Equinox for model definitions and Optax for optimization. The repository provides:

  • A minimal and clean implementation of HER in JAX;
  • Reproducible scripts and results;
  • A Colab Notebook for direct experimentation.

Don't hesitate to check the code: https://github.com/jeertmans/HER-with-JAX.

Let me know if you have any questions, feedback, or recommendations!