[oehpr]

I wonder if there might be a way to map all these using t-SNE to discrete grid locations? Maybe even an autoencoder. I'd love to see what features it could pick out.

I don't see their data set though. hmmm.

maybe I'll just have to crawl it on my own if I want to do it.

[lgvld]

You can use t-SNE (or even better: UMAP or one of its variation) to create a 2D points cloud, and then use something like RasterFairy [1] to map 2D positions to the cells a grid. It usually works well.

[1] https://github.com/Quasimondo/RasterFairy

[yboris]

side note: instead of t-SNE consider UMAP - provides better results (and it's much faster) https://github.com/lmcinnes/umap