identified problem with svm: need probability=true for AUROC

model_selection/test_models.py

@@ -8,7 +8,7 @@ import pandas as pd
 import numpy as np
 from xgboost import XGBClassifier
 from sklearn.metrics import confusion_matrix
-from sklearn.metrics import f1_score, make_scorer, precision_score, recall_score, accuracy_score
+from sklearn.metrics import f1_score, make_scorer, precision_score, recall_score, accuracy_score, roc_auc_score, average_precision_score
 from sklearn.ensemble import RandomForestClassifier, BaggingClassifier, AdaBoostClassifier
 from sklearn.neural_network import MLPClassifier
 from sklearn.svm import SVC
@@ -49,7 +49,7 @@ def get_tuned_models(group_id, method_id):
             "AB" : AdaBoostClassifier(**{'learning_rate': 1.9189147333140566, 'n_estimators': 131, 'algorithm': 'SAMME'}), 
             "XGB": XGBClassifier(**{'learning_rate': 0.22870029177880222, 'max_depth': 8, 'n_estimators': 909}),
             "LR" : LogisticRegression(**{'solver': 'lbfgs', 'penalty': None, 'max_iter': 1000}), 
-            "SVM" : SVC(**{'C': 0.9872682949695772, 'kernel': 'linear', 'max_iter':1000}), 
+            #"SVM" : SVC(**{'C': 0.9872682949695772, 'kernel': 'linear', 'max_iter':1000}), 
             "MLP" : MLPClassifier(**{'activation': 'identity', 'hidden_layer_sizes': 122, 'learning_rate': 'invscaling', 'max_iter':500})
             }
         # 1.2) Trained with original dataset and cost-sensitive learning
@@ -60,7 +60,7 @@ def get_tuned_models(group_id, method_id):
             "Bagging": BaggingClassifier(**{'max_features': 1.0, 'max_samples': 1.0, 'n_estimators': 15, 'warm_start': False, 'estimator': DecisionTreeClassifier(class_weight='balanced')}),
             "AB": AdaBoostClassifier(**{'learning_rate': 0.8159074545140872, 'n_estimators': 121, 'algorithm': 'SAMME', 'estimator': DecisionTreeClassifier(class_weight='balanced')}),
             "LR": LogisticRegression(**{'solver': 'lbfgs', 'penalty': None, 'max_iter': 1000, 'class_weight': 'balanced'}),
-            "SVM": SVC(**{'C': 1.5550524351360953, 'kernel': 'linear', 'max_iter': 1000, 'class_weight': 'balanced'}),
+            #"SVM": SVC(**{'C': 1.5550524351360953, 'kernel': 'linear', 'max_iter': 1000, 'class_weight': 'balanced'}),
             }
         # 1.3) Trained with oversampled training dataset
         elif method_id == 2:
@@ -71,7 +71,7 @@ def get_tuned_models(group_id, method_id):
             "AB" : AdaBoostClassifier(**{'learning_rate': 1.6590924545876917, 'n_estimators': 141, 'algorithm': 'SAMME'}), 
             "XGB": XGBClassifier(**{'learning_rate': 0.26946295284728783, 'max_depth': 7, 'n_estimators': 893}),
             "LR" : LogisticRegression(**{'solver': 'lbfgs', 'penalty': 'l2', 'max_iter': 1000}), 
-            "SVM" : SVC(**{'C': 1.676419306008229, 'kernel': 'poly', 'max_iter':1000}), 
+            #"SVM" : SVC(**{'C': 1.676419306008229, 'kernel': 'poly', 'max_iter':1000}), 
             "MLP" : MLPClassifier(**{'activation': 'relu', 'hidden_layer_sizes': 116, 'learning_rate': 'invscaling', 'max_iter':500})
             }
         # 1.4) Trained with undersampled training dataset
@@ -83,7 +83,7 @@ def get_tuned_models(group_id, method_id):
             "AB" : AdaBoostClassifier(**{'learning_rate': 1.6996764264041269, 'n_estimators': 93, 'algorithm': 'SAMME'}), 
             "XGB": XGBClassifier(**{'learning_rate': 0.26480707899668926, 'max_depth': 7, 'n_estimators': 959}),
             "LR" : LogisticRegression(**{'solver': 'lbfgs', 'penalty': None, 'max_iter': 1000}), 
-            "SVM" : SVC(**{'C': 1.1996501173654208, 'kernel': 'poly', 'max_iter':1000}), 
+            #"SVM" : SVC(**{'C': 1.1996501173654208, 'kernel': 'poly', 'max_iter':1000}), 
             "MLP" : MLPClassifier(**{'activation': 'relu', 'hidden_layer_sizes': 131, 'learning_rate': 'constant', 'max_iter':500})
             }
     # 2. POST
@@ -97,7 +97,7 @@ def get_tuned_models(group_id, method_id):
             "AB" : AdaBoostClassifier(**{'learning_rate': 1.7806904141367559, 'n_estimators': 66, 'algorithm': 'SAMME'}), 
             "XGB": XGBClassifier(**{'learning_rate': 0.21889089898592098, 'max_depth': 6, 'n_estimators': 856}),
             "LR" : LogisticRegression(**{'solver': 'lbfgs', 'penalty': None, 'max_iter': 1000}), 
-            "SVM" : SVC(**{'C': 1.9890638540240584, 'kernel': 'linear', 'max_iter':1000}), 
+            #"SVM" : SVC(**{'C': 1.9890638540240584, 'kernel': 'linear', 'max_iter':1000}), 
             "MLP" : MLPClassifier(**{'activation': 'logistic', 'hidden_layer_sizes': 112, 'learning_rate': 'constant', 'max_iter':500})
             }
         # 2.2) Trained with original dataset and cost-sensitive learning
@@ -108,7 +108,7 @@ def get_tuned_models(group_id, method_id):
             "Bagging": BaggingClassifier(**{'max_features': 1.0, 'max_samples': 0.8, 'n_estimators': 11, 'warm_start': True, 'estimator': DecisionTreeClassifier(class_weight='balanced')}),
             "AB": AdaBoostClassifier(**{'learning_rate': 1.7102248217141944, 'n_estimators': 108, 'algorithm': 'SAMME', 'estimator': DecisionTreeClassifier(class_weight='balanced')}),
             "LR": LogisticRegression(**{'solver': 'lbfgs', 'penalty': None, 'max_iter': 1000, 'class_weight': 'balanced'}),
-            "SVM": SVC(**{'C': 1.1313840454519628, 'kernel': 'sigmoid', 'max_iter': 1000, 'class_weight': 'balanced'})
+            #"SVM": SVC(**{'C': 1.1313840454519628, 'kernel': 'sigmoid', 'max_iter': 1000, 'class_weight': 'balanced'})
             }
         # 2.3) Trained with oversampled training dataset
         elif method_id == 2:
@@ -131,7 +131,7 @@ def get_tuned_models(group_id, method_id):
             "AB" : AdaBoostClassifier(**{'learning_rate': 1.836659462701278, 'n_estimators': 138, 'algorithm': 'SAMME'}), 
             "XGB": XGBClassifier(**{'learning_rate': 0.2517946893282251, 'max_depth': 4, 'n_estimators': 646}),
             "LR" : LogisticRegression(**{'solver': 'lbfgs', 'penalty': 'l2', 'max_iter': 1000}), 
-            "SVM" : SVC(**{'C': 1.8414678085000697, 'kernel': 'linear', 'max_iter':1000}), 
+            #"SVM" : SVC(**{'C': 1.8414678085000697, 'kernel': 'linear', 'max_iter':1000}), 
             "MLP" : MLPClassifier(**{'activation': 'relu', 'hidden_layer_sizes': 76, 'learning_rate': 'constant', 'max_iter':500})
             }
     return tuned_models
@@ -188,9 +188,10 @@ if __name__ == "__main__":
         'TN':TN_scorer, 
         'FN':FN_scorer, 
         'FP':FP_scorer, 
-        'TP':TP_scorer
+        'TP':TP_scorer,
+        'AUROC': make_scorer(roc_auc_score, needs_threshold=True),  # AUROC requires decision function or probability outputs
+        'AUPRC': make_scorer(average_precision_score, needs_proba=True)  # AUPRC requires probability outputs
         } 
-        # AUROC and AUPRC (plot?)
     method_names = {
         0: "ORIG",
         1: "ORIG_CW",