# Fixing precision problem in discriminant calculation

## Summary

This MR introduces the following changes

- Ensure full precision is used for the probability values (coming from the dumper) which are used to calculate the discriminant score. If in half precision, this can lead to NaNs in some Zprime jets.
- Setting the default value for the
`frac_min`

to 0 in the rejection per fraction calculation (helpful when running with more than 3 classes and you want to exclude one) - Fixing an issue in
`add_variables`

check in the evaluate_model script. With the new dataclass based configs, this is always an empty list and therefore the is not None check will always pass. Checking now also the length of the list.