Feature/random test set

refactoring-ai / predicting-refactoring-ml

Refactoring recommendation via ML

MIT License

28 stars 8 forks source link

Feature/random test set #200

Closed jan-gerling closed 4 years ago

jan-gerling commented 4 years ago

Fix Bugs:

fixed incorrect output formatting for test scores
fix minor bug in QueryBuilder test_get_level_refactorings_count

Features:

multiple validation sets for the model evaluation
- specify validation sets in config
move DROP_PROCESS_AND_AUTHORSHIP_METRICS column specification to logs
adopt output to generate logs for multiple validation sets
every call to log also prints

Other

Refactor preprocessing retrieve_labelled_instances

jan-gerling commented 4 years ago

You can't shuffle x and y separately.

We have to use the scaler generated in the training to build the validation dataset. Otherwise, things won't match.

Both major issues are fixed.