Open rajeeja opened 1 year ago
Another approach that can be added - maybe a little difficult -- to use uno_preprocess.py
to create a .h5 file and use the config parameter --use-exported and load that hdf5 file for train and infer. That'd be much less intrusive for the original model.
Map files from the
Uno
way of doing things toIMPROVE
way: UNO: https://web.cels.anl.gov/projects/IMPROVE_FTP/candle/public/benchmarks/Pilot1/combo/IMPROVE: https://ftp.mcs.anl.gov/pub/candle/public/improve/IMP_data/
Replace uno_data CombinedDataGenerator, CombinedDataLoader, DataFeeder with improve equivalents.
Work on Uno_IMPROVE folder for this work. The goal is to be able to do cross-study like the other IMPROVE models.