MNIST example

[stites]

A good place to start getting familiar with the API that doesn't require access to a GPU (unlike #64) is porting the MNIST example from pytorch/examples and doing a small write-up. All the other haskell libraries are doing it and everyone loves writing about it.

[ocramz]

Here's a encoding/decoding library for the MNIST storage format : https://hackage.haskell.org/package/mnist-idx

[AdLucem]

Working on this, but with the fashion-mnist dataset (https://github.com/zalandoresearch/fashion-mnist).

[stites]

Looking forward to it! Entirely optional, but I noticed that you are familiar with dh-core, so you might want to think about adding fashion-mnist to datasets. There's a streaming interface for attoparsec parsers here: DataHaskell/dh-core#31 , and you can use the CIFAR-10 parser as a guideline (see Numeric.Datasets.CIFAR10).

Definitely scope-creep so don't feel like you need to do it at all, but if you're parsing MNIST formats from binary then it could be an excellent contribution to dh-core.

Also please be as noisy as possible on this issue if you run into any blockers.

[austinvhuang]

Closing this issue with #196