iancovert/persist

Speed?

Opened this issue · 1 comments

cnk113 commented

Hello,

I'm running PERSIST on my scRNA-seq dataset, however I have >100K cells and I noticed training taking quite a while. I was wondering what I can do to speedup training? For reference I have 48gb GPU memory.

Best,
Chang

iancovert commented

Hi Chang, thanks for checking out the package and apologies for my slow response. I suspect that the model trains slowly for your dataset because of the large # samples (>100K) and the large # of training epochs. The number of training epochs could most likely be significantly lower without compromising the results, but it's hard to say a priori how much lower.

We're currently working on revisions to the paper/package, and one of the updates will be removing # epochs as a hyperparameter (because it's a bit costly to tune). I expect that will make PERSIST quite a bit faster for your data, but it will take some time (possibly a couple weeks) before our changes are finalized.

Ian