Data creation, model training, model validation: time benchmark

Test data

• source: https://zenodo.org/records/3265189
• original data: DES_Ia-0001_HEAD.FITS, DES_Ia-0001_PHOT.FITS, DES_NONIa-0001_HEAD.FITS, DES_NONIa-0001_PHOT.FITS
• size:
  • DES_Ia-0001: 108 MB
  • DES_NONIa-000: 124.4 MB
• number of SN objects: 46901
  • DES_Ia-0001: 21558
  • DES_NONIa-000: 25343

Test setup

• use Ngarrgu Tindebeek supercomputers
• --mem=10g --cpus-per-task=2

Time benchmark

Original code

1. Data creation (python run.py --data --dump_dir ../data/dump_data --raw_dir ../data/raw)

• total running time: 37.7 s
• total cpu time: 15.9 s

2. rnn train (python run.py --train_rnn --dump_dir ../data/dump_data --nb_epoch=5)

• total running time: 76.6 s
• total cpu time: 74.4 s

3. rnn validate (python run.py --validate_rnn --dump_dir ../data/dump_data)

• total running time: 7.2 s
• total cpu time: 7.2 s