PMBio / deeprvat

Other

16 stars 1 forks source link

Feature: Optimizations and bug fixes for training #42

Closed bfclarke closed 6 months ago

bfclarke commented 7 months ago

What

Bug fix: Subset samples correctly

Optimizations to training:

Sample subsetting now happens in the training_dataset rule
Uses in-memory instead of on-disk Zarr arrays to speed up data loading
Trains multiple models in parallel to improve GPU utilization

Testing

Run pytest
Run the snakemake pipeline on the example data
Association testing on real data to verify model is training correctly