Skip to content

Version 0.3.7

Latest

Choose a tag to compare

@felicitas215 felicitas215 released this 01 Jul 13:48

Changelog Version 0.3.7

  • fixed wrong HelixerPost default for min_coding_length in Helixer.py, affects all predictions generated through 1-step inference via Helixer.py
  • --predict-phase is now the default for every model, every script specifying --predict-phase for HybridModel.py needs to remove this argument as it no longer exists (issue #103)
  • multiple small bug fixes (no influence on prediction results)
  • add reproducibility notes and ensure determinism during inference using same hard- and software for GPUs using non-deterministic cuDNN algorithms
  • remove nni support as the package is discontinued/archived
  • add import, integration and training tests through pytest and add/improve test data
  • fix fully intergenic subsequences calculation, reported percentage of fully intergenic samples will now be correct for any newly created HDF5 files, old files will still report the wrong amount (this has no effect on prediction accuracy or training and is just used as information for the user)
  • add more safeguards against HDF5 read/write race-conditions
  • logging is more consistent now; Python version and a list of GPUs that tensorflow will use are logged now
  • add_ngs_coverage.py's --dataset-prefix now defaults to rnaseq
  • bump versions of minor packages to address dependabot pull requests
  • additional information: add the updated link to the Zenodo models since the mammal model was added for manual download, add the approximate Helixer installation time for inexperienced users