torchnca

torchnca copied to clipboard

nan/divide by zero issues in loss function with sparse data input

6

Sparse input matrix (30k cells x 6000 genes, single-cell sequencing data, with 35 labels for the cells) produces nan's in the loss function during training. Adding a small epsilon, 1e-12,...

tarachari3

Can you also add how good is it compared to tSNE?

2

njanirudh

The t-SNE is running with default values. We may need to play around with the hyperparams to get better results.

njanirudh

Multilabel case

1

Hey kevin, I tried implementing this on JAX as a learning exercise (as you saw on twitter!) and thought it might be useful to have a multilabel case, so I'll...

ljmartin