def test_load_model(self): pca = Pca(name='TestPca') pca.load(self.base_path) h5py.File.assert_called_once_with( os.path.join(self.base_path, 'processors/TestPca.h5'), 'r') if sys.version_info[0] < 3: pickle.loads.assert_called_once_with('Model') else: pickle.loads.assert_called_once_with('Model', encoding='latin1') self.assertEquals('Model', pca._model)
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']))))