Internal function of predict_trait_MET_cv()
.
Plots are done at the CV scheme level from the , which means that:
If CV0 is evaluated, the plot shows the 40 most important variables according to the predicted element (i.e. site, year or environment).
For CV1 and CV2, variable importance plots are based on a average of the importance of each feature over all training/test splits.
Variable importance can be calculated based on model agnostic approaches (permutation-based methods, like for stacking_reg_1
or DL_reg
), or
on model-specific methods (gain metric for GBDT methods xgb_reg
).
plot_results_vip_cv( fitting_all_splits, cv_type, cv0_type, path_folder, nb_folds_cv1, repeats_cv1, nb_folds_cv2, repeats_cv2 )
fitting_all_splits | a |
---|---|
cv_type | A |
cv0_type | cv0_type A |
path_folder | a |
nb_folds_cv1 | A |
repeats_cv1 | A |
nb_folds_cv2 | A |
repeats_cv2 | A |
A variable importance plot is saved in the path_folder
. No specific object returned.
Cathy C. Westhues cathy.jubin@uni-goettingen.de