Skip to content

Latest commit

 

History

History
16 lines (10 loc) · 700 Bytes

README.md

File metadata and controls

16 lines (10 loc) · 700 Bytes
Raw

RadioGalaxies-BBB

Code for variational inference for radio galaxy classification based on the Bayes by Backprop algorithm.

Training

  • Run bbb_ensemble.py to train 10 variational inference models with different random seeds and random shuffling between training and validation dataset. Uses wandb logger.

Configs for experiments:

  • Use configt.txt to run vi without data agumentation
  • Use config_augment.txt to run vi with data agumentation

Evaluation

  • Run pred_analysis.py to calculate test error and uncertainty calibration error for predictive entropy for 10 experimental runs.
  • Run analysis.py to calculate energy scores for different test sets (MiraBest, MIGHTEE, GalaxyMNIST).