def decision_tree_performance(X_train, y_train, X_test, y_test, eval_method,
threshold_list):
'''
This function generate the performance matrix of decision tree model
Inputs:
X_train, y_train, X_test, y_test: Traing and testing dataframes
eval_method: evaluation method used in the function
threshold_list: list of the percentiles of the population for
extracting thresholds from the predicted probability
Returns: performance matrix
'''
columns = ["criterion", "splitter", "max_depth", "best_thres", "best_eval"]
m_performance = pd.DataFrame(columns=columns)
for crit in DECISION_TREE_HP["criterion"]:
for split in DECISION_TREE_HP["splitter"]:
for m_depth in DECISION_TREE_HP["max_depth"]:
dt = decision_tree.classifier_settings_dt(\
criterion = crit, splitter=split, max_depth=m_depth)
y_predp = decision_tree.train_decision_tree(X_train, \
y_train, X_test, dt)[1]
best_thres, best_eval = eva.best_threshold(y_test, \
y_predp, eval_method, threshold_list)
output = {"criterion":[crit], "splitter":[split], "max_depth":[m_depth], \
"best_thres":[best_thres], "best_eval":[best_eval]}
sub_p = pd.DataFrame(data=output)
m_performance = pd.concat([m_performance, sub_p], join="inner")
m_performance["method"] = "decision_tree"
return m_performance
def bagging_performance(X_train, y_train, X_test, y_test, base_method,
eval_method, threshold_list):
'''
This function generate the performance matrix of bagging model
Inputs:
X_train, y_train, X_test, y_test: Traing and testing dataframes
base_method: the method to be passed in for boosting
eval_method: evaluation method used in the function
threshold_list: list of the percentiles of the population for
extracting thresholds from the predicted probability
Returns: performance matrix
'''
columns = ["n_estimators", "best_thres", "best_eval"]
m_performance = pd.DataFrame(columns=columns)
for n in BAGGING_HP["n_estimators"]:
bag = bagging.classifier_settings_bagging(n_estimators=n,
base_estimator=base_method)
y_predp = bagging.train_bagging(X_train, y_train, X_test, bag)[1]
best_thres, best_eval = eva.best_threshold(y_test, \
y_predp, eval_method, threshold_list)
output = {
"n_estimators": [n],
"best_thres": [best_thres],
"best_eval": [best_eval]
}
sub_p = pd.DataFrame(data=output)
m_performance = pd.concat([m_performance, sub_p], join="inner")
m_performance["method"] = "bagging"
return m_performance
def logistics_performance(X_train, y_train, X_test, y_test, eval_method,
threshold_list):
'''
This function generate the performance matrix of logistics model
Inputs:
X_train, y_train, X_test, y_test: Traing and testing dataframes
eval_method: evaluation method used in the function
threshold_list: list of the percentiles of the population for
extracting thresholds from the predicted probability
Returns: performance matrix
'''
columns = ["penalty", "C", "best_thres", "best_eval"]
m_performance = pd.DataFrame(columns=columns)
for pan in LOG_HP["penalty"]:
for c in LOG_HP["C"]:
log_m = logistic.classifier_settings_log(penalty=pan, C=c)
y_predp = logistic.train_logistics(X_train, y_train, X_test,
log_m)[1]
best_thres, best_eval = eva.best_threshold(y_test, \
y_predp, eval_method, threshold_list)
output = {"penalty":[pan], "C":[c], "best_thres":[best_thres], \
"best_eval":[best_eval]}
sub_p = pd.DataFrame(data=output)
m_performance = pd.concat([m_performance, sub_p], join="inner")
m_performance["method"] = "logistics"
return m_performance
def svm_performance(X_train, y_train, X_test, y_test, eval_method,
threshold_list):
'''
This function generate the performance matrix of svm model
Inputs:
X_train, y_train, X_test, y_test: Traing and testing dataframes
eval_method: evaluation method used in the function
threshold_list: list of the percentiles of the population for
extracting thresholds from the predicted probability
Returns: performance matrix
'''
columns = ["kernel", "C", "best_thres", "best_eval"]
m_performance = pd.DataFrame(columns=columns)
for ker in SVM_HP["kernel"]:
for c in SVM_HP["C"]:
svm_m = svm.classifier_settings_svm(kernel=ker, C=c)
y_predp = svm.train_svm(X_train, y_train, X_test, svm_m)[1]
best_thres, best_eval = eva.best_threshold(y_test, \
y_predp, eval_method, threshold_list)
output = {"kernel":[ker], "C":[c], "best_thres":[best_thres], \
"best_eval":[best_eval]}
sub_p = pd.DataFrame(data=output)
m_performance = pd.concat([m_performance, sub_p], join="inner")
m_performance["method"] = "svm"
return m_performance
def knn_performance(X_train, y_train, X_test, y_test, eval_method,
threshold_list):
'''
This function generate the performance matrix of knn model
Inputs:
X_train, y_train, X_test, y_test: Traing and testing dataframes
eval_method: evaluation method used in the function
threshold_list: list of the percentiles of the population for
extracting thresholds from the predicted probability
Returns: performance matrix
'''
columns = ["weights", "metric", "n_neighbors", "best_thres", "best_eval"]
m_performance = pd.DataFrame(columns=columns)
for wgt in KNN_HP["weights"]:
for met in KNN_HP["metric"]:
for n in KNN_HP["n_neighbors"]:
knn_m = knn.classifier_settings_knn(\
weights=wgt, metric=met, n_neighbors=n)
y_predp = knn.train_knn(X_train, y_train, X_test, knn_m)[1]
best_thres, best_eval = eva.best_threshold(y_test, \
y_predp, eval_method, threshold_list)
output = {"weight":[wgt], "metric":[met], "n_neighbors":[n], \
"best_thres":[best_thres], "best_eval":[best_eval]}
sub_p = pd.DataFrame(data=output)
m_performance = pd.concat([m_performance, sub_p], join="inner")
m_performance["method"] = "knn"
return m_performance
def random_forest_performance(X_train, y_train, X_test, y_test, eval_method,
threshold_list):
'''
This function generate the performance matrix of random forest model
Inputs:
X_train, y_train, X_test, y_test: Traing and testing dataframes
eval_method: evaluation method used in the function
threshold_list: list of the percentiles of the population for
extracting thresholds from the predicted probability
Returns: performance matrix
'''
columns = [
"criterion", "n_estimators", "max_depth", "best_thres", "best_eval"
]
m_performance = pd.DataFrame(columns=columns)
for crit in RANDOM_FOREST_HP["criterion"]:
for n in RANDOM_FOREST_HP["n_estimators"]:
for m_depth in RANDOM_FOREST_HP["max_depth"]:
rf = decision_tree.classifier_settings_rf(\
criterion = crit, n_estimators=n, max_depth=m_depth)
y_predp = decision_tree.train_random_forest(X_train, \
y_train, X_test, rf)[1]
best_thres, best_eval = eva.best_threshold(y_test, \
y_predp, eval_method, threshold_list)
output = {"criterion":[crit], "n_estimators":[n], "max_depth":[m_depth], \
"best_thres":[best_thres], "best_eval":[best_eval]}
sub_p = pd.DataFrame(data=output)
m_performance = pd.concat([m_performance, sub_p], join="inner")
m_performance["method"] = "random_forset"
return m_performance