search

mlangguth89 / downscaling_benchmark

6 stars 0 forks source link

Fix change in error calculation in run_feature_importance-method #12

Closed mlangguth89 closed 1 month ago

mlangguth89 commented 5 months ago

The relative change in error should not be calculated against the averaged error, but against its timeseries.
In particular, the following lines must be corrected:


    score_data = pd.read_csv(score_file)
    ref_score = score_data[f"{score_name}_mean"].mean()

    rel_changes = feature_scores / ref_score
´´´
where `feature_scores` provide the score timeseries with the perurbed feature.