An open API service indexing awesome lists of open source software.

https://github.com/pkhungurn/shion

My personal machine learning experimentation, designed to work with PyTorch.
https://github.com/pkhungurn/shion

experimentation machine-learning python pytorch training

Last synced: about 2 months ago
JSON representation

My personal machine learning experimentation, designed to work with PyTorch.

Awesome Lists containing this project

README

          

# shion

`shion` (シオン) is my personal machine learning experimentation, designed to work with PyTorch.

## What does it do?

On top of machine learning frameworks such as [TensorFlow](https://www.tensorflow.org), [PyTorch](https://pytorch.org),
and [JAX](https://github.com/google/jax), you need something like [fast.ai](https://www.fast.ai/) or
[PyTorch Lightning](https://www.pytorchlightning.ai/) to make training and inference easy. Something like this should
give you abstractions for:

* the evaluation of loss functions,
* the invocation of optimizers,
* the saving of snapshots,
* the resumption from previous training run,
* the logging of training and validation losses, and
* the visualization of your model’s outputs once in a while.

`shion` provides them in such a way that is hell-bent on separation of concerns and customization. I have stuck with
the library in the past since 2019 without making many changes to the core abstractions. The caveat, though, is that
it is quite hard to use even for myself.

## Installation

You can just copy the `src\shion` directory to your codebase, or you can use the following tools.

### Pip

```
pip install git+https://github.com/pkhungurn/shion.git
```

### Poetry

```
poetry add git+https://github.com/pkhungurn/shion.git
```

## Update History

* (2022/01/05) v0.1.1: Allow specifying dependencies of training tasks.
* (2022/01/05) v0.1.0: First release.