Skip to content

v1.2.0: robustness improve on the LoRa finetuning

Latest

Choose a tag to compare

@ruochiz ruochiz released this 04 Jun 21:40
· 18 commits to main since this release

Multiple bug fixes:

  • the error where seq2print_tfbs might ended up using only the last model, instead of averaging over five folds

Multiple improvement:

  • support custom region names, if there is a column "region_names" in the region provided, those would be used to name results instead of the standard chr1:0-10000 etc.
  • better verbose
  • chromvar also chunk over TF axis for larger datasets
  • More robust LoRA training with better coverage warming stages