def main():
data = load_breast_cancer()
features = data['data']
labels = data['target']
features_train, features_test, labels_train, labels_test = train_test_split(features,
labels,
test_size=0.2,
random_state=42)
x_train = {
'features': features_train,
'labels': labels_train
}
x_test = {
'features': features_test,
'labels': labels_test
}
save_path = '/tmp/pipeline_example'
intermediate_features_path = os.path.join(save_path, 'features')
pipeline_train = Pipeline('Pca_and_linear_svc_pipeline',[Pca(n_components=0.95), LinearSvc(C=1.0)])
print ('Training pipeline description: ', pipeline_train)
y_train = pipeline_train.fit_run(x_train)
print('AUC for training set: {}'.format(str(roc_auc_score(y_train['labels'], y_train['scores']))))
pipeline_train.save(save_path)
pca = Pca()
pca.load(save_path)
svc = LinearSvc()
svc.load(save_path)
pipeline_test = Pipeline('Pca_and_linear_svc_pipeline', [pca,
PipelineSaver(intermediate_features_path,
'PCA_test_features'),
svc])
pipeline_test.load(save_path)
print('Test pipeline description: ', pipeline_test)
y_test = pipeline_test.run(x_test)
print('AUC for training set: {}'.format(str(roc_auc_score(y_test['labels'], y_test['scores']))))
def test_run_pipeline_saver(self):
reduced_features = RecursiveFeatureElimination(n_features=1)
pipeline = Pipeline('name_pipeline', [
reduced_features,
PipelineSaver('dim_reduction', 'dim_reduction')
])
pipeline.fit(self.X)
pipeline.save('/tmp/test_rfe_pipeline_saver')
loaded_pipeline = Pipeline(
'name_pipeline',
[PipelineLoader('dim_reduction', 'dim_reduction')])
loaded_pipeline.load('/tmp/test_rfe_pipeline_saver')
Y = loaded_pipeline.run(self.X)
np.testing.assert_equal(
Y['features'],
np.array(([[0.2], [0.1], [0.1], [0.2], [0.9], [0.8]])))
np.testing.assert_equal(Y['labels'], np.array([0, 0, 0, 0, 1, 1]))
np.testing.assert_equal(Y['access_ids'], np.array([0, 0, 1, 1, 2, 2]))
np.testing.assert_equal(Y['indices'], np.array(
[2])) # indices of the selected features