diff --git a/model_selection/test_models.py b/model_selection/test_models.py index a3a83a4dbdf0701176a091f285528d6c0f2dd53c..cb79d9c86181dfaea0ae70b5bba394f48969091a 100644 --- a/model_selection/test_models.py +++ b/model_selection/test_models.py @@ -18,6 +18,7 @@ from sklearn.metrics import RocCurveDisplay, roc_curve from sklearn.metrics import PrecisionRecallDisplay, precision_recall_curve import matplotlib.pyplot as plt from sklearn.metrics import confusion_matrix, ConfusionMatrixDisplay +import ast # String to dictionary # -------------------------------------------------------------------------------------------------------- # Reading test data @@ -71,103 +72,47 @@ def read_test_data(): # Returning tuned models for each situation # -------------------------------------------------------------------------------------------------------- -def get_tuned_models(group_id, method_id): - # 1. PRE - if group_id == 0: - # 1.1) Trained with original dataset - if method_id == 0: - tuned_models = { - "DT" : DecisionTreeClassifier(**{'splitter': 'best', 'max_features': 'sqrt', 'criterion': 'entropy'}), - "RF" : RandomForestClassifier(**{'criterion': 'entropy', 'max_features': 'sqrt', 'n_estimators': 123}), - "Bagging" : BaggingClassifier(**{'max_features': 1.0, 'max_samples': 0.8, 'n_estimators': 13, 'warm_start': False}), - "AB" : AdaBoostClassifier(**{'learning_rate': 1.8473150336970519, 'n_estimators': 96, 'algorithm': 'SAMME'}), - "XGB": XGBClassifier(**{'learning_rate': 0.21528982071549305, 'max_depth': 6, 'n_estimators': 804}), - "LR" : LogisticRegression(**{'solver': 'lbfgs', 'penalty': 'l2','max_iter': 1000}), - "SVM" : SVC(**{'C': 1.051871311397777, 'kernel': 'linear', 'max_iter':1000, 'probability': True}), - "MLP" : MLPClassifier(**{'activation': 'identity', 'hidden_layer_sizes': 78, 'learning_rate': 'constant','max_iter':500}) - } - # 1.2) Trained with original dataset and cost-sensitive learning - elif method_id == 1: - tuned_models = { - "DT": DecisionTreeClassifier(**{'splitter': 'best', 'max_features': 'log2', 'criterion': 'entropy', 'class_weight': 'balanced'}), - "RF": RandomForestClassifier(**{'criterion': 'entropy', 'max_features': 'sqrt', 'n_estimators': 238, 'class_weight': 'balanced'}), - "Bagging": BaggingClassifier(**{'max_features': 1.0, 'max_samples': 0.8, 'n_estimators': 22, 'warm_start': False, 'estimator': DecisionTreeClassifier(class_weight='balanced')}), - "AB": AdaBoostClassifier(**{'learning_rate': 1.7136783954287846, 'n_estimators': 99, 'algorithm': 'SAMME', 'estimator': DecisionTreeClassifier(class_weight='balanced')}), - "LR": LogisticRegression(**{'solver': 'lbfgs', 'penalty': 'l2', 'max_iter': 1000, 'class_weight': 'balanced'}), - "SVM": SVC(**{'C': 1.480857958217729, 'kernel': 'linear', 'max_iter': 1000, 'class_weight': 'balanced', 'probability': True}), - } - # 1.3) Trained with oversampled training dataset - elif method_id == 2: - tuned_models = { - "DT" : DecisionTreeClassifier(**{'splitter': 'best', 'max_features': 'sqrt', 'criterion': 'log_loss'}), - "RF" : RandomForestClassifier(**{'criterion': 'gini', 'max_features': 'sqrt', 'n_estimators': 121}), - "Bagging" : BaggingClassifier(**{'max_features': 1.0, 'max_samples': 1.0, 'n_estimators': 22, 'warm_start': True}), - "AB" : AdaBoostClassifier(**{'learning_rate': 1.4640913091426446, 'n_estimators': 145, 'algorithm': 'SAMME'}), - "XGB": XGBClassifier(**{'learning_rate': 0.19621698151985992, 'max_depth': 7, 'n_estimators': 840}), - "LR" : LogisticRegression(**{'solver': 'lbfgs', 'penalty': 'l2', 'max_iter': 1000}), - "SVM" : SVC(**{'C': 1.590799972846728, 'kernel': 'poly', 'max_iter':1000, 'probability': True}), - "MLP" : MLPClassifier(**{'activation': 'relu', 'hidden_layer_sizes': 112, 'learning_rate': 'constant', 'max_iter':500}) - } - # 1.4) Trained with undersampled training dataset - elif method_id == 3: - tuned_models = { - "DT" : DecisionTreeClassifier(**{'splitter': 'best', 'max_features': 'sqrt', 'criterion': 'log_loss'}), - "RF" : RandomForestClassifier(**{'criterion': 'gini', 'max_features': 'sqrt', 'n_estimators': 148}), - "Bagging" : BaggingClassifier(**{'max_features': 1.0, 'max_samples': 0.8, 'n_estimators': 24, 'warm_start': True}), - "AB" : AdaBoostClassifier(**{'learning_rate': 1.7970533619575801, 'n_estimators': 122, 'algorithm': 'SAMME'}), - "XGB": XGBClassifier(**{'learning_rate': 0.13148624656904934, 'max_depth': 9, 'n_estimators': 723}), - "LR" : LogisticRegression(**{'solver': 'sag', 'penalty': 'l2', 'max_iter': 1000}), - "SVM" : SVC(**{'C': 1.383651513577477, 'kernel': 'poly', 'max_iter':1000, 'probability': True}), - "MLP" : MLPClassifier(**{'activation': 'relu', 'hidden_layer_sizes': 89, 'learning_rate': 'invscaling', 'max_iter':500}) - } - # 2. POST - else: - # 2.1) Trained with original dataset - if method_id == 0: - tuned_models = { - "DT" : DecisionTreeClassifier(**{'splitter': 'best', 'max_features': 'sqrt', 'criterion': 'log_loss'}), - "RF" : RandomForestClassifier(**{'criterion': 'entropy', 'max_features': 'sqrt', 'n_estimators': 120}), - "Bagging" : BaggingClassifier(**{'max_features': 1.0, 'max_samples': 0.8, 'n_estimators': 38, 'warm_start': True}), - "AB" : AdaBoostClassifier(**{'learning_rate': 1.9069394544838472, 'n_estimators': 121, 'algorithm': 'SAMME'}), - "XGB": XGBClassifier(**{'learning_rate': 0.24787889985627387, 'max_depth': 4, 'n_estimators': 956}), - "LR" : LogisticRegression(**{'solver': 'lbfgs', 'penalty': 'l2'}), - "SVM" : SVC(**{'C': 1.7965537393241109, 'kernel': 'linear', 'max_iter':1000, 'probability': True}), - "MLP" : MLPClassifier(**{'activation': 'relu', 'hidden_layer_sizes': 147, 'learning_rate': 'invscaling', 'max_iter':500}) - } - # 2.2) Trained with original dataset and cost-sensitive learning - elif method_id == 1: - tuned_models = { - "DT": DecisionTreeClassifier(**{'splitter': 'best', 'max_features': 'sqrt', 'criterion': 'gini', 'class_weight': 'balanced'}), - "RF": RandomForestClassifier(**{'criterion': 'entropy', 'max_features': 'sqrt', 'n_estimators': 138, 'class_weight': 'balanced'}), - "Bagging": BaggingClassifier(**{'max_features': 1.0, 'max_samples': 1.0, 'n_estimators': 66, 'warm_start': True, 'estimator': DecisionTreeClassifier(class_weight='balanced')}), - "AB": AdaBoostClassifier(**{'learning_rate': 1.92541653518023, 'n_estimators': 114, 'algorithm': 'SAMME', 'estimator': DecisionTreeClassifier(class_weight='balanced')}), - "LR": LogisticRegression(**{'solver': 'lbfgs', 'penalty': 'l2', 'max_iter': 1000, 'class_weight': 'balanced'}), - "SVM": SVC(**{'C': 0.8395104850983046, 'kernel': 'linear', 'max_iter': 1000, 'class_weight': 'balanced', 'probability': True}) - } - # 2.3) Trained with oversampled training dataset - elif method_id == 2: - tuned_models = { - "DT" : DecisionTreeClassifier(**{'splitter': 'best', 'max_features': 'log2', 'criterion': 'entropy'}), - "RF" : RandomForestClassifier(**{'criterion': 'gini', 'max_features': 'sqrt', 'n_estimators': 118}), - "Bagging" : BaggingClassifier(**{'max_features': 1.0, 'max_samples': 1.0, 'n_estimators': 56, 'warm_start': False}), - "AB" : AdaBoostClassifier(**{'learning_rate': 1.5933610622176648, 'n_estimators': 114, 'algorithm': 'SAMME'}), - "XGB": XGBClassifier(**{'learning_rate': 0.059934879882855396, 'max_depth': 9, 'n_estimators': 660}), - "LR" : LogisticRegression(**{'solver': 'lbfgs', 'penalty': 'l2', 'max_iter': 1000}), - "SVM" : SVC(**{'C': 1.2237930722499044, 'kernel': 'poly', 'max_iter':1000, 'probability': True}), - "MLP" : MLPClassifier(**{'activation': 'identity', 'hidden_layer_sizes': 134, 'learning_rate': 'invscaling', 'max_iter':500}) - } - # 2.4) Trained with undersampled training dataset - elif method_id == 3: - tuned_models = { - "DT" : DecisionTreeClassifier(**{'splitter': 'best', 'max_features': 'log2', 'criterion': 'log_loss'}), - "RF" : RandomForestClassifier(**{'criterion': 'gini', 'max_features': 'sqrt', 'n_estimators': 151}), - "Bagging" : BaggingClassifier(**{'max_features': 1.0, 'max_samples': 1.0, 'n_estimators': 20, 'warm_start': False}), - "AB" : AdaBoostClassifier(**{'learning_rate': 1.6523810056317618, 'n_estimators': 89, 'algorithm': 'SAMME'}), - "XGB": XGBClassifier(**{'learning_rate': 0.18430397856234193, 'max_depth': 4, 'n_estimators': 956}), - "LR" : LogisticRegression(**{'solver': 'lbfgs', 'penalty': 'l2', 'max_iter': 1000}), - "SVM" : SVC(**{'C': 1.1807459108651588, 'kernel': 'linear', 'max_iter':1000, 'probability': True}), - "MLP" : MLPClassifier(**{'activation': 'identity', 'hidden_layer_sizes': 55, 'learning_rate': 'constant', 'max_iter':500}) - } +def get_tuned_models(group_str, method_str): + + # Read sheet corresponding to group and method with tuned models and their hyperparam + tuned_models_df = pd.read_excel("./output_hyperparam/hyperparamers.xlsx",sheet_name=f"{group_str}_{method_str}") + # Mapping from model abbreviations to sklearn model classes + model_mapping = { + 'DT': DecisionTreeClassifier, + 'RF': RandomForestClassifier, + 'Bagging': BaggingClassifier, + 'AB': AdaBoostClassifier, + 'XGB': XGBClassifier, + 'LR': LogisticRegression, + 'SVM': SVC, + 'MLP': MLPClassifier + } + tuned_models = {} + # Iterate through each row of the DataFrame + for index, row in tuned_models_df.iterrows(): + model_name = row[0] + # Read dictionary + parameters = ast.literal_eval(row['Parameters']) + # Add extra parameters + if model_name == 'AB': + parameters['algorithm'] = 'SAMME' + elif model_name == 'LR': + parameters['max_iter'] = 1000 + elif model_name == 'SVM': + parameters['max_iter'] = 1000 + parameters['probability'] = True + elif model_name == "MLP": + parameters['max_iter'] = 500 + # Add class_weight argument for cost-sensitive learning method + if 'CW' in method_str: + if model_name == 'Bagging' or model_name == 'AB': + parameters['estimator'] = DecisionTreeClassifier(class_weight='balanced') + else: + parameters['class_weight'] = 'balanced' + # Fetch class + model_class = model_mapping[model_name] + # Initialize model + tuned_models[model_name] = model_class(**parameters) return tuned_models # -------------------------------------------------------------------------------------------------------- @@ -242,12 +187,12 @@ if __name__ == "__main__": X_test = data_dic['X_test_' + group] y_test = data_dic['y_test_' + group] for j, method in enumerate(['', '', 'over_', 'under_']): - print(f"{group}-{method}") + print(f"{group}-{method_names[j]}") # Get train dataset based on group and method X_train = data_dic['X_train_' + method + group] y_train = data_dic['y_train_' + method + group] # Get tuned models for this group and method - models = get_tuned_models(group_id=i, method_id=j) + models = get_tuned_models(group, method_names[j]) # Scores df scores_df = pd.DataFrame(index=models.keys(), columns=scorings.keys()) # Create a figure for all models in this group-method @@ -292,6 +237,6 @@ if __name__ == "__main__": with pd.ExcelWriter('./test_results/testing_tuned_models.xlsx') as writer: for sheet_name, data in scores_sheets.items(): data.to_excel(writer, sheet_name=sheet_name) - # -------------------------------------------------------------------------------------------------------- +# --------------------------------------------------------------------------------------------------------