To study the influence of a single (hyper-)parameter, I use validation_curve:
.... grid_result = grid_obj.fit(X_train1, y_train1) grid_best = grid_result.best_estimator_ train_score, val_score = validation_curve(grid_best , X_train1 , y_train1 , param_name = 'learning_rate' , param_range = param_range , cv=5 , scoring="neg_mean_squared_error" , verbose = 1 )
param_name can hold only a single variable. Thus, though when e.g. already performing a hyperparameter search through
gridsearchcv or something similar, I have to fit/CV all single parameters in
There is probably a smarter way, or?