https://github.com/game4all/mnist-from-scratch

🔍 Handwritten digit classifier neural network written and trained on MNIST dataset from scratch using Zig programming language.
https://github.com/game4all/mnist-from-scratch

machine-learning ml mnist-classification wasm webassembly ziglang

Last synced: 28 days ago
JSON representation

🔍 Handwritten digit classifier neural network written and trained on MNIST dataset from scratch using Zig programming language.

• Host: GitHub
• URL: https://github.com/game4all/mnist-from-scratch
• Owner: Game4all
• License: mit
• Created: 2024-08-01T08:35:04.000Z (5 months ago)
• Default Branch: master
• Last Pushed: 2024-09-06T18:12:55.000Z (4 months ago)
• Last Synced: 2024-10-16T19:20:45.759Z (3 months ago)
• Topics: machine-learning, ml, mnist-classification, wasm, webassembly, ziglang
• Language: Zig
• Homepage: https://game4all.github.io/mnist-from-scratch
• Size: 14.7 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        


    
`mnist-from-scratch`

    A handwritten digit classsifier neural network written and trained from scratch in pure zig

    


    Uses brainz library for model training.

    




![UI Screenshot](assets/ui.png)

## Getting it up running

This requires Zig **0.13.0** to work.

1. Clone the git repository

2. Download and extract the MNIST dataset into the repository root.

2. Run `zig build`

3. Deployable WASM + HTML file are at `zig-out/web/public`

## Training the model locally

You can train the model locally by running `zig build train`. By default the training will use the model checkpoints in `src/model.bin`. You can start from scratch by deleting the checkpoints and running the command again.

## License and acknowledgements

This project is licensed under **MIT licence**.

Uses the MNIST dataset (https://yann.lecun.com/exdb/mnist/)