def dataset3_svc_sgd(dataset):
x_original = dataset.loc[:, dataset.columns != 'labels']
labels = dataset['labels']
fselect = FeatureSelection(dataset, x_original, labels)
model_lsvc = LinearSVC(C=0.1, penalty="l1", dual=False)
#model= linearsvs
X_fit_model, X_transf_model,column_selected,feature_importances,feature_importances_DF,dataset_features= \
fselect.select_from_model_feature_elimination(model=model_lsvc)
ml = MachineLearning(X_transf_model, labels, classes=['pos', 'neg'])
filename = 'crmapp/ml_models/dataset3_all_'
print('best model sgd')
best_sgd_model = ml.train_best_model('sgd')
print('score_test_set_sgd')
print(ml.score_testset(best_sgd_model))
print('roc curve')
ml.plot_roc_curve(best_sgd_model)
pickle.dump(best_sgd_model, open(filename + 'sgd_model_15092020.sav',
'wb'))
def dataset3_univariate_svm(dataset):
x_original = dataset.loc[:, dataset.columns != 'labels']
labels = dataset['labels']
fselect = FeatureSelection(dataset, x_original, labels)
model_lsvc = LinearSVC(C=0.1, penalty="l1", dual=False)
# #**Select KBest**
# #KBest com *mutual info classif*
X_fit_univariate, X_transf_univariate,column_selected,scores,dataset_features = \
fselect.univariate(score_func=mutual_info_classif, mode='k_best', param=500)
ml = MachineLearning(X_transf_univariate, labels, classes=['pos', 'neg'])
filename = 'crmapp/ml_models/dataset3_all_'
#tests models
print('best model svm')
best_svm_model = ml.train_best_model('svm')
print('score_test_set_svm')
print(ml.score_testset(best_svm_model))
print('roc curve')
ml.plot_roc_curve(best_svm_model)
def dataset1_tree_svm(dataset):
x_original = dataset.loc[:, dataset.columns != 'labels']
labels = dataset['labels']
fselect = FeatureSelection(dataset, x_original, labels)
model_tree = ExtraTreesClassifier(n_estimators=50)
# Select from model
#model= Tree classifier. 50 estiamtors
X_fit_model, X_transf_model,column_selected,feature_importances,feature_importances_DF,dataset_features= \
fselect.select_from_model_feature_elimination(model=model_tree)
ml = MachineLearning(X_transf_model, labels, classes=['pos', 'neg'])
filename = 'crmapp/ml_models/dataset1_all_'
#tests models
print('best model svm')
best_svm_model = ml.train_best_model('svm')
print('score_test_set_svm')
print(ml.score_testset(best_svm_model))
print('roc curve')
ml.plot_roc_curve(best_svm_model)
pickle.dump(best_svm_model, open(filename + 'svm_model_15092020.sav',
'wb'))
def test_feature_selection(dataset):
x_original = dataset.loc[:, dataset.columns != 'labels']
labels = dataset['labels']
fselect = FeatureSelection(dataset, x_original, labels)
"""
# #**Select KBest**
# #KBest com *mutual info classif*
X_fit_univariate, X_transf_univariate,column_selected,scores,dataset_features = \
fselect.univariate(score_func=mutual_info_classif, mode='k_best', param=1000)
#**Select Percentile
#Percentile with *f classif*
X_fit_univariate, X_transf_univariate,column_selected,scores,dataset_features = \
fselect.univariate(score_func=f_classif, mode='percentile', param=0.6)
# Select only the features with p value inferior to 0.015
X_fit_univariate, X_transf_univariate,column_selected,scores,dataset_features \
= fselect.univariate(score_func=f_classif, mode='fpr', param=0.05)
#shape of transformed dataset
print(X_transf_univariate.shape)
#columns selected by high score
print(fselect.features_scores(x_original,scores,column_selected, False))
#**SRecursive feature elimination
#estimator=SVC kernel=linear with 5 cross validation
X_fit_rfe, X_transf_rfe,column_selected,ranking,dataset_features= \
fselect.recursive_feature_elimination(cross_validation=True,cv=5)
#shape of transformed dataset
print(X_transf_rfe.shape)
#columns selected names
print(dataset.columns[column_selected])
#scores
score_methods(x_original,X_transf_rfe,labels)
#
"""
# Select from model
# L1-based feature selection f linear_model.LogisticRegression/svm.LinearSVC for classification
# With SVMs and logistic-regression, the parameter C controls the sparsity: the smaller C the fewer features selected.
model_lsvc = LinearSVC(C=1, penalty="l1", dual=False)
model_lr = LogisticRegression(C=0.1, penalty="l2", dual=False)
model_tree = ExtraTreesClassifier(n_estimators=50)
"""
# Select from model
#model= Tree classifier. 50 estiamtors
X_fit_model, X_transf_model,column_selected,feature_importances,feature_importances_DF,dataset_features= \
fselect.select_from_model_feature_elimination(model=model_tree)
#model= logistic regression
X_fit_model, X_transf_model,column_selected,feature_importances,feature_importances_DF,dataset_features= \
fselect.select_from_model_feature_elimination( model=model_lr)
"""
#model= linearsvs
X_fit_model, X_transf_model,column_selected,feature_importances,feature_importances_DF,dataset_features= \
fselect.select_from_model_feature_elimination(model=model_lsvc)
"""
print('original shape', dataset.shape)
print('reduce shape', fselect.dataset.shape)
print('dataset reduced with column names\n', fselect.dataset.head(3))
print('feature importances\n',feature_importances_DF)
print('scores')
score_methods(x_original,X_transf_model,labels)
print(fselect.dataset)
"""
return fselect.dataset