MNIST digit classification

[mchav]

Data: https://www.kaggle.com/datasets/hojjatk/mnist-dataset
Goal: Classify hand drawn digits

[nstepp]

I would like to work on this one. It would also be easily extendable to more advanced datasets, which I think would be a good opportunity to show how using Haskell has benefits in that situation. I'm imagining taking an MNIST model and then explain how you would add layers and change shapes for a different problem.

[mchav]

Hasktorch has a pretty good selection of MNIST implementations.

https://github.com/hasktorch/hasktorch/tree/master/examples

For the iris dataset we took the core logic of the Hasktorch and put a lot of data science looking ceremony around it so feel free to do the same.

But maybe you could also improve the example. I think we're eventually thinking to make a library that wraps Hasktorch. So thinking through the API in addition to the functionality would be great.