Feature Importance use unaggregated results

Feature Importance

SKill scores of bootstrapped predictions for feature importance are currently aggregated to mean for each station and later plotted as box-and-whisker.

Target

Refactore code so that not the mean for each station but the real skill scores (e.g. N=20 per station) are used to create the feature importance plot. This can enhance the understanding of feature importance, because more data is plotted. Currently it only indicates as a mean value which variable is important for each individual station.

ToDo

• remove aggregation of skill scores (will retain the boots dimension)
• be able to plot this data