Skip to content

C
covid_analysis
Commit 2d351899 authored by Joaquin Torres's avatar Joaquin Torres
Browse files
Options

automated reading the parameters for tuned models

parent 6674b724
Show whitespace changes
Inline Side-by-side
Showing with 45 additions and 100 deletions
model_selection/test_models.py
View file @ 2d351899
...@@ -18,6 +18,7 @@ from sklearn.metrics import RocCurveDisplay, roc_curve ...@@ -18,6 +18,7 @@ from sklearn.metrics import RocCurveDisplay, roc_curve
from sklearn.metrics import PrecisionRecallDisplay, precision_recall_curve from sklearn.metrics import PrecisionRecallDisplay, precision_recall_curve
import matplotlib.pyplot as plt import matplotlib.pyplot as plt
from sklearn.metrics import confusion_matrix, ConfusionMatrixDisplay from sklearn.metrics import confusion_matrix, ConfusionMatrixDisplay
import ast # String to dictionary
# -------------------------------------------------------------------------------------------------------- # --------------------------------------------------------------------------------------------------------
# Reading test data # Reading test data
...@@ -71,103 +72,47 @@ def read_test_data(): ...@@ -71,103 +72,47 @@ def read_test_data():
# Returning tuned models for each situation # Returning tuned models for each situation
# -------------------------------------------------------------------------------------------------------- # --------------------------------------------------------------------------------------------------------
def get_tuned_models(group_id, method_id): def get_tuned_models(group_str, method_str):
# 1. PRE
if group_id == 0: # Read sheet corresponding to group and method with tuned models and their hyperparam
# 1.1) Trained with original dataset tuned_models_df = pd.read_excel("./output_hyperparam/hyperparamers.xlsx",sheet_name=f"{group_str}_{method_str}")
if method_id == 0: # Mapping from model abbreviations to sklearn model classes
tuned_models = { model_mapping = {
"DT" : DecisionTreeClassifier(**{'splitter': 'best', 'max_features': 'sqrt', 'criterion': 'entropy'}), 'DT': DecisionTreeClassifier,
"RF" : RandomForestClassifier(**{'criterion': 'entropy', 'max_features': 'sqrt', 'n_estimators': 123}), 'RF': RandomForestClassifier,
"Bagging" : BaggingClassifier(**{'max_features': 1.0, 'max_samples': 0.8, 'n_estimators': 13, 'warm_start': False}), 'Bagging': BaggingClassifier,
"AB" : AdaBoostClassifier(**{'learning_rate': 1.8473150336970519, 'n_estimators': 96, 'algorithm': 'SAMME'}), 'AB': AdaBoostClassifier,
"XGB": XGBClassifier(**{'learning_rate': 0.21528982071549305, 'max_depth': 6, 'n_estimators': 804}), 'XGB': XGBClassifier,
"LR" : LogisticRegression(**{'solver': 'lbfgs', 'penalty': 'l2','max_iter': 1000}), 'LR': LogisticRegression,
"SVM" : SVC(**{'C': 1.051871311397777, 'kernel': 'linear', 'max_iter':1000, 'probability': True}), 'SVM': SVC,
"MLP" : MLPClassifier(**{'activation': 'identity', 'hidden_layer_sizes': 78, 'learning_rate': 'constant','max_iter':500}) 'MLP': MLPClassifier
}
# 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 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: else:
# 2.1) Trained with original dataset parameters['class_weight'] = 'balanced'
if method_id == 0: # Fetch class
tuned_models = { model_class = model_mapping[model_name]
"DT" : DecisionTreeClassifier(**{'splitter': 'best', 'max_features': 'sqrt', 'criterion': 'log_loss'}), # Initialize model
"RF" : RandomForestClassifier(**{'criterion': 'entropy', 'max_features': 'sqrt', 'n_estimators': 120}), tuned_models[model_name] = model_class(**parameters)
"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})
}
return tuned_models return tuned_models
# -------------------------------------------------------------------------------------------------------- # --------------------------------------------------------------------------------------------------------
...@@ -242,12 +187,12 @@ if __name__ == "__main__": ...@@ -242,12 +187,12 @@ if __name__ == "__main__":
X_test = data_dic['X_test_' + group] X_test = data_dic['X_test_' + group]
y_test = data_dic['y_test_' + group] y_test = data_dic['y_test_' + group]
for j, method in enumerate(['', '', 'over_', 'under_']): 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 # Get train dataset based on group and method
X_train = data_dic['X_train_' + method + group] X_train = data_dic['X_train_' + method + group]
y_train = data_dic['y_train_' + method + group] y_train = data_dic['y_train_' + method + group]
# Get tuned models for this group and method # 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
scores_df = pd.DataFrame(index=models.keys(), columns=scorings.keys()) scores_df = pd.DataFrame(index=models.keys(), columns=scorings.keys())
# Create a figure for all models in this group-method # Create a figure for all models in this group-method
...@@ -292,6 +237,6 @@ if __name__ == "__main__": ...@@ -292,6 +237,6 @@ if __name__ == "__main__":
with pd.ExcelWriter('./test_results/testing_tuned_models.xlsx') as writer: with pd.ExcelWriter('./test_results/testing_tuned_models.xlsx') as writer:
for sheet_name, data in scores_sheets.items(): for sheet_name, data in scores_sheets.items():
data.to_excel(writer, sheet_name=sheet_name) data.to_excel(writer, sheet_name=sheet_name)
# -------------------------------------------------------------------------------------------------------- # --------------------------------------------------------------------------------------------------------
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment