Skip to content

Latest commit

 

History

History
67 lines (46 loc) · 1.57 KB

README.md

File metadata and controls

67 lines (46 loc) · 1.57 KB
Raw

Haiku Utilities and Framework: huf

Code style: black

  • Various utilities for working with haiku
  • A minimal keras-inspired framework for supervised learning
  • gin configurable CLI

Installation

After installing jax

pip install dm-haiku
git clone https://github.com/jackd/huf.git
cd huf
pip install -e .

Quick Start

See the mnist example for a simple classification example. To use the CLI,

cd examples/mnist
python -m huf huf_config/fit.gin configs/base.gin

You can also experiment with tweaked configurations:

python -m huf huf_config/fit.gin configs/base.gin --bindings='
batch_size = 32
epochs = 12
dropout_rate = 0.6
'

Note this is equivalent to

python -m huf huf_config/fit.gin configs/tweaked.gin

Projects using HUF

  • grax: graph networks with jax

Pre-commit

This package uses pre-commit to ensure commits meet minimum criteria. To Install, use

pip install pre-commit
pre-commit install

This will ensure git hooks are run before each commit. While it is not advised to do so, you can skip these hooks with

git commit --no-verify -m "commit message"

TODO

  • Document everything
  • Seperate jax data library that focuses just on data, like tf.data (dax?, jata?)