Skip to content

Update scaling and writing

Samuel Van Stroud requested to merge svanstro/update-scaling-writing into master

Summary

This MR introduces the following changes

  • Standardise scale dict format between jets and tracks (take the track format)
  • Allow concat_jet_tracks to be a list of jet variables in order to concat only a subset of jet variables
  • Store each track label in it's own dataset, allowing them to be more easily accessed (and allow for the definition of each one to be stored in the metadata in the future)
  • Default prepared path prepared_samples/->prepared/ to match preprocessed/
  • Default scale dict name PFlow-scale_dict.json -> scale_dict.json since PFlow is not really helpful and is clear from other info
  • Save track origin weights to the scale dict
  • Default to 16bit precision of final train files (#212)
  • Various cleanup

Relates to the following issues

Conformity

Edited by Samuel Van Stroud

Merge request reports