Skip to content

Add fit-based targets to hyperparameter optimization

Emanuele Roberto Nocera requested to merge future_test_as_hyperopt_reward into master

Created by: scarlehoff

This is a continuation for #1313 and #1314

After this PR, targets for the hyperoptimizaton based on the fits coming from the folds can be used. As a first example I've added the distance between folds. I will also add the future tests as a metric in some way (still not sure what the metric should be).

kfold:
  target: fit_distance

Obviously by default you have one single replica to the CV, but you can add now the right flags to the runcard:

genrep: true
parallel_models: true
same_trvl_per_replica: true

And off you go:

n3fit runcards/Basic_hyperopt.yml 1 -r 5 --hyperopt 10

Merge request reports