fitmodels.py

from sklearn.metrics import mean_absolute_error
from eli5.sklearn import PermutationImportance

mae = pd.DataFrame(columns = ['train', 'test'])
fi = pd.DataFrame(columns = feature_names)

for model_name in list(models.keys()):

    # fit model
    models[model_name].fit(X_trn, y_trn)
    
    # compute mean absolute error of model in train and test set
    mae.loc[model_name,:] = [mean_absolute_error(y_trn, models[model_name].predict(X_trn)), mean_absolute_error(y_tst, models[model_name].predict(X_tst))]
    
    # compute feature importances of model
    try:
        feature_importances_ = models[model_name].feature_importances_
    except:
        feature_importances_ = PermutationImportance(models[model_name], cv = 'prefit', n_iter = 3).fit(X_trn, y_trn).feature_importances_
    fi.loc[model_name, :] = feature_importances_ / feature_importances_.sum()
    
fi.fillna(0, inplace = True)