def run_best_mlp(ds_id, all_labels, train_features, train_labels, val_features, val_labels, test_features, test_labels):
# These hyper parameters were found using the GridSearchCV algorithm
best_mlp_clf = MLPClassifier(hidden_layer_sizes=(60, 80), activation='tanh', solver='adam')
# Define hyper params to search
# parameter_space = {
# 'hidden_layer_sizes': [(50, 80), (20, 20, 20)],
# 'activation': ['identity', 'tanh', 'logistic', 'relu'],
# 'solver': ['adam', 'sgd']
# }
# mlp_clf_with_gs = GridSearchCV(best_mlp_clf, parameter_space, n_jobs=-1, cv=5)
# Train
best_mlp_clf = best_mlp_clf.fit(train_features, train_labels)
# print('Best parameters found:\n', mlp_clf_with_gs.best_params_)
if val_features is not None and val_labels is not None:
# Validation
pred_labels = best_mlp_clf.predict(val_features)
# Evaluation of Validation data
clf_evaluation(val_labels, pred_labels, all_labels, 'Best-MLP_val.csv')
else:
# Test
pred_labels = best_mlp_clf.predict(test_features)
# Evaluation of Test data
clf_evaluation(test_labels, pred_labels, all_labels, 'Best-MLP-' + ds_id + '.csv')
def run_base_dt(ds_id, all_labels, train_features, train_labels, val_features, val_labels, test_features, test_labels):
base_dt_clf = DecisionTreeClassifier(criterion='entropy')
# Train
base_dt_clf = base_dt_clf.fit(train_features, train_labels)
if val_features is not None and val_labels is not None:
# Validation
pred_labels = base_dt_clf.predict(val_features)
# Evaluation of Validation data
clf_evaluation(val_labels, pred_labels, all_labels, 'Base-DT_val.csv')
else:
# Test
pred_labels = base_dt_clf.predict(test_features)
# Evaluation of Test data
clf_evaluation(test_labels, pred_labels, all_labels, 'Base-DT-' + ds_id + '.csv')
def run_naive_bayes(ds_id, all_labels, train_features, train_labels,
val_features, val_labels, test_features, test_labels):
clf = GaussianNB()
# Train
clf.fit(train_features, train_labels)
if val_features is not None and val_labels is not None:
# Validation
pred_labels = clf.predict(val_features)
# Evaluation of Validation data
clf_evaluation(val_labels, pred_labels, all_labels, 'GNB_val.csv')
else:
# Test
pred_labels = clf.predict(test_features)
# Evaluation of Test data
clf_evaluation(test_labels, pred_labels, all_labels,
'GNB-' + ds_id + '.csv')
def run_perceptron(ds_id, all_labels, train_features, train_labels,
val_features, val_labels, test_features, test_labels):
perceptron_clf = Perceptron()
# Train
perceptron_clf.fit(train_features, train_labels)
if val_features is not None and val_labels is not None:
# Validation
pred_labels = perceptron_clf.predict(val_features)
# Evaluation of Validation data
clf_evaluation(val_labels, pred_labels, all_labels, 'PER_val.csv')
else:
# Test
pred_labels = perceptron_clf.predict(test_features)
# Evaluation of Test data
clf_evaluation(test_labels, pred_labels, all_labels,
'PER-' + ds_id + '.csv')
def run_best_dt(ds_id, all_labels, train_features, train_labels, val_features,
val_labels, test_features, test_labels):
# These hyper parameters were found using the GridSearchCV algorithm
base_dt_clf = DecisionTreeClassifier(criterion='entropy',
max_depth=None,
class_weight=None,
min_impurity_decrease=0,
min_samples_split=2)
# Define hyper params to search
# parameter_space = {
# 'criterion': ['gini', 'entropy'],
# 'max_depth': [10, None],
# 'min_samples_split': [8, 10, 20, 30, 40],
# 'min_impurity_decrease': [0, 0.0002, 0.0003, 0.0001],
# 'class_weight': [None, 'balanced']
# }
# dt_clf_with_gs = GridSearchCV(base_dt_clf, parameter_space, n_jobs=-1, cv=5)
# Train
base_dt_clf = base_dt_clf.fit(train_features, train_labels)
# print('Best parameters found:\n', dt_clf_with_gs.best_params_)
if val_features is not None and val_labels is not None:
# Validation
pred_labels = base_dt_clf.predict(val_features)
# Evaluation of Validation data
clf_evaluation(val_labels, pred_labels, all_labels, 'Best-DT_val.csv')
else:
# Test
pred_labels = base_dt_clf.predict(test_features)
# Evaluation of Test data
clf_evaluation(test_labels, pred_labels, all_labels,
'Best-DT-' + ds_id + '.csv')
def run_base_mlp(ds_id, all_labels, train_features, train_labels, val_features,
val_labels, test_features, test_labels):
clf_base_mlp = MLPClassifier(hidden_layer_sizes=100,
activation='logistic',
solver='sgd')
# Train
clf_base_mlp.fit(train_features, train_labels)
if val_features is not None and val_labels is not None:
# Validation
pred_labels = clf_base_mlp.predict(val_features)
# Evaluation of Validation data
clf_evaluation(val_labels, pred_labels, all_labels, 'Base-MLP_val.csv')
else:
# Test
pred_labels = clf_base_mlp.predict(test_features)
# Evaluation of Test data
clf_evaluation(test_labels, pred_labels, all_labels,
'Base-MLP-' + ds_id + '.csv')