def test_generic_estimator_with_distribution_param(): # Note that in normal situations (release build), init_connection_args can be omitted # otherwise, it should be set to the first H2O element in the pipeline pipeline = Pipeline([('svd', TruncatedSVD(n_components=3, random_state=seed)), ('estimator', H2OGradientBoostingEstimator( distribution='bernoulli', seed=seed, init_connection_args=init_connection_args))]) data = _get_data(format='numpy') assert isinstance(data.X_train, np.ndarray) pipeline.fit(data.X_train, data.y_train) preds = pipeline.predict(data.X_test) assert isinstance(preds, np.ndarray) assert preds.shape == (len(data.X_test), ) probs = pipeline.predict_proba(data.X_test) assert probs.shape == (len(data.X_test), 2) assert np.allclose(np.sum(probs, axis=1), 1.), "`predict_proba` didn't return probabilities" score = pipeline.score(data.X_test, data.y_test) assert isinstance(score, float) skl_score = accuracy_score(data.y_test, preds) assert abs(score - skl_score) < 1e-6 scores['generic_estimator_with_distribution_param'] = score
def test_params_are_correctly_passed_to_underlying_estimator(): estimator = H2OGradientBoostingEstimator(seed=seed) estimator.set_params(max_depth=10, learn_rate=0.5) estimator.model_id = "dummy" assert estimator.estimator is None estimator._make_estimator() # normally done when calling `fit` real_estimator = estimator.estimator assert real_estimator parms = real_estimator._parms assert real_estimator.seed == parms['seed'] == seed assert real_estimator.max_depth == parms['max_depth'] == 10 assert real_estimator.learn_rate == parms['learn_rate'] == 0.5 assert real_estimator._id == parms['model_id'] == "dummy" assert real_estimator.training_frame == parms['training_frame'] is None
def test_all_params_can_be_set_as_properties(): pipeline = Pipeline([('standardize', H2OScaler()), ('pca', H2OPCA()), ('estimator', H2OGradientBoostingEstimator())]) pipeline.named_steps.standardize.center = True pipeline.named_steps.standardize.scale = False pipeline.named_steps.pca.k = 2 pipeline.named_steps.pca.seed = seed pipeline.named_steps.estimator.ntrees = 20 pipeline.named_steps.estimator.max_depth = 5 pipeline.named_steps.estimator.seed = seed params = pipeline.get_params() assert isinstance(params['standardize'], H2OScaler) assert params['standardize__center'] is True assert params['standardize__scale'] is False assert isinstance(params['pca'], H2OPCA) assert params['pca__k'] == 2 assert params['pca__seed'] == seed assert isinstance(params['estimator'], H2OGradientBoostingEstimator) assert params['estimator__ntrees'] == 20 assert params['estimator__max_depth'] == 5 assert params['estimator__seed'] == seed
def test_all_params_can_be_set_using_set_params(): pipeline = Pipeline([('standardize', H2OScaler()), ('pca', H2OPCA()), ('estimator', H2OGradientBoostingEstimator())]) pipeline.set_params(standardize__center=True, standardize__scale=False, pca__k=2, pca__seed=seed, estimator__ntrees=20, estimator__max_depth=5, estimator__seed=seed) assert isinstance(pipeline.named_steps.standardize, H2OScaler) assert pipeline.named_steps.standardize.center is True assert pipeline.named_steps.standardize.scale is False assert isinstance(pipeline.named_steps.pca, H2OPCA) assert pipeline.named_steps.pca.k == 2 assert pipeline.named_steps.pca.seed == seed assert isinstance(pipeline.named_steps.estimator, H2OGradientBoostingEstimator) assert pipeline.named_steps.estimator.ntrees == 20 assert pipeline.named_steps.estimator.max_depth == 5 assert pipeline.named_steps.estimator.seed == seed
def test_all_params_are_accessible_as_properties(): pipeline = Pipeline([('standardize', H2OScaler(center=True, scale=False)), ('pca', H2OPCA(k=2, seed=seed)), ('estimator', H2OGradientBoostingEstimator(ntrees=20, max_depth=5, seed=seed))]) assert isinstance(pipeline.named_steps.standardize, H2OScaler) assert pipeline.named_steps.standardize.center is True assert pipeline.named_steps.standardize.scale is False assert isinstance(pipeline.named_steps.pca, H2OPCA) assert pipeline.named_steps.pca.k == 2 assert pipeline.named_steps.pca.seed == seed assert isinstance(pipeline.named_steps.estimator, H2OGradientBoostingEstimator) assert pipeline.named_steps.estimator.ntrees == 20 assert pipeline.named_steps.estimator.max_depth == 5 assert pipeline.named_steps.estimator.seed == seed # also the ones that were not set explicitly assert pipeline.named_steps.pca.max_iterations is None assert pipeline.named_steps.estimator.learn_rate is None
def test_all_params_are_visible_in_get_params(): pipeline = Pipeline([('standardize', H2OScaler(center=True, scale=False)), ('pca', H2OPCA(k=2, seed=seed)), ('estimator', H2OGradientBoostingEstimator(ntrees=20, max_depth=5, seed=seed))]) params = pipeline.get_params() assert isinstance(params['standardize'], H2OScaler) assert params['standardize__center'] is True assert params['standardize__scale'] is False assert isinstance(params['pca'], H2OPCA) assert params['pca__k'] == 2 assert params['pca__seed'] == seed assert isinstance(params['estimator'], H2OGradientBoostingEstimator) assert params['estimator__ntrees'] == 20 assert params['estimator__max_depth'] == 5 assert params['estimator__seed'] == seed # also the ones that were not set explicitly assert params['pca__max_iterations'] is None assert params['estimator__learn_rate'] is None
def test_generic_estimator_with_distribution_param(): # Note that in normal situations (release build), init_connection_args can be omitted # otherwise, it should be set to the first H2O element in the pipeline pipeline = Pipeline([('standardize', StandardScaler()), ('pca', PCA(n_components=2, random_state=seed)), ('estimator', H2OGradientBoostingEstimator( distribution='gaussian', seed=seed, init_connection_args=init_connection_args))]) data = _get_data(format='numpy') assert isinstance(data.X_train, np.ndarray) pipeline.fit(data.X_train, data.y_train) preds = pipeline.predict(data.X_test) assert isinstance(preds, np.ndarray) assert preds.shape == (len(data.X_test), ) score = pipeline.score(data.X_test, data.y_test) assert isinstance(score, float) skl_score = r2_score(data.y_test, preds) assert abs(score - skl_score) < 1e-6 scores['generic_estimator_with_distribution_param'] = score