如何找出在 Python 中网格搜索中选择的最佳 model 的精度、召回率、特异性和 F1 分数？

Question

This is the Grid Selection step for my model. I am able to find out the best accuracy score and best parameters of that selected model but I want to find out precision, recall, specificity & F1 score of the best model which is selected in Grid Search

from sklearn.model_selection import GridSearchCV
parameters = [{'n_estimators': [100, 200],
               'max_features': ['auto', 'sqrt', None], 
               'max_depth': [10, 20, 30, None], 
               'criterion': ['gini', 'entropy'],
               'min_samples_split':  [5, 10,15], 'min_samples_leaf': [1,4,6], 
               'bootstrap': [True, False]}]

grid_search = GridSearchCV(estimator = classifier,
                           param_grid = parameters,
                           scoring = 'accuracy',
                           cv = 5,
                           n_jobs = -1) #n_jobs to optimise grid search process
grid_search.fit(X_train, Y_train)
best_accuracy = grid_search.best_score_
best_parameters = grid_search.best_params_
print("\n")
print("Results for Grid Search")
print("Best Accuracy: {:.2f} %".format(best_accuracy*100))
print("Best Parameters:", best_parameters)

Answer 1

GridSearchCV enables passing a list of scoring functions, provided that you specify a single scoring function to which the best parameters will be chosen to refit the model.

grid_search = GridSearchCV(
    estimator=classifier,
    param_grid=parameters,
    scoring=['accuracy', 'f1', 'precision', 'recall'],
    refit="accuracy",  # Or any other value from `scoring` list
)

Then you can access the entire cross validation results in the cv_results_ attribute. It can be easier to first pack the cv_results_ as a pandas DataFrame and then access the row corresponding to best_index_ attribute:

cv_results = pd.DataFrame(grid_search.cv_results_)
best_model_results = cv.results.loc[grid_search.best_index_]

The you get a Series indexed by score names, so you can access, for example, "mean_test_recall" , "mean_test_f1" , etc.

Note that, for your question, specificity is not a built-in scoring name , but you can also supply GridSearchCV with custom metrics , so it's possible to pass a callable calculating it (say by using confusion matrix )