def test_lightgbm_classifier(self):
""" test converting lightgbm model
"""
gbm = lgb.LGBMClassifier(boosting_type='gbdt',
num_leaves=31,
max_depth=6,
learning_rate=0.1,
n_estimators=10,
objective='multiclass')
gbm.fit(self.data.X,
self.data.y,
eval_metric='logloss',
feature_name=self.data.features)
model_output = os.path.join(self.data.base_path,
"lightgbm_classifier.txt")
gbm.booster_.save_model(model_output)
self.assertTrue(os.path.exists(model_output))
pmml_output = os.path.join(self.data.base_path,
"lightgbm_classifier.pmml")
ppmml.to_pmml(model_input=model_output,
pmml_output=pmml_output,
model_type='lightgbm',
options={
'compact': True,
'target-name': 'y'
})
self.assertTrue(os.path.exists(pmml_output))
# validate pmml file
data_output = os.path.join(self.data.base_path,
"lightgbm_classifier.csv")
ppmml.predict(pmml_output, self.data.test_data_input, data_output)
self.assertTrue(os.path.exists(data_output))
def test_linear_classifier(self):
""" test converting LinearClassifer model
"""
algorithm_name = "linear_classifer"
model_output = os.path.join(self.data.base_path,
"{}".format(algorithm_name))
classifier = LinearClassifier(
feature_columns=self._iris_dnn_features(),
n_classes=3,
optimizer=tf.train.AdamOptimizer,
config=self.estimator_conf)
self._generate_tf_model(classifier, model_output)
self.assertTrue(os.path.exists(model_output))
pmml_output = os.path.join(self.data.base_path,
"{}.pmml".format(algorithm_name))
ppmml.to_pmml(model_input=model_output,
pmml_output=pmml_output,
model_type='tensorflow')
self.assertTrue(os.path.exists(pmml_output))
# validate pmml file
data_output = os.path.join(self.data.base_path,
"{}.csv".format(algorithm_name))
ppmml.predict(pmml_output, self.data.test_data_input, data_output)
self.assertTrue(os.path.exists(data_output))
def test_r_converter(self):
""" test converting r model
"""
pmml_output = os.path.join(self.base_path, "r_model.pmml")
ppmml.to_pmml(model_input=self.r_model_path,
pmml_output=pmml_output,
model_type='r')
self.assertTrue(os.path.exists(pmml_output))
def _to_pmml_and_validate(self, algo_name):
""" to pmml and validate
Args:
algo_name: string, the algorithm name
"""
model_output = os.path.join(self.base_path, "{}_model".format(algo_name))
schema_output = os.path.join(self.base_path, "{}.json".format(algo_name))
pmml_output = os.path.join(self.base_path, "{}.pmml".format(algo_name))
ppmml.to_pmml(model_output, pmml_output, schema_input=schema_output, model_type="spark")
self.assertTrue(os.path.exists(pmml_output))
# validate pmml file
if not os.path.exists(self.test_data_path):
logging.info("generate test data: {}".format(self.test_data_path))
self._generate_test_data(self.test_data_path)
data_output = os.path.join(self.base_path, "{}.csv".format(algo_name))
ppmml.predict(pmml_output, self.test_data_path, data_output)
self.assertTrue(os.path.exists(data_output))
def _to_pmml_and_validate(self, estimator, algo_name):
""" to pmml and validate
Args:
estimator: sklearn estimator
algo_name: string, the algorithm name
"""
model_output = os.path.join(self.data.base_path,
"{}.pkl.z".format(algo_name))
pmml_output = os.path.join(self.data.base_path,
"{}.pmml".format(algo_name))
joblib.dump(estimator, model_output, compress=9)
ppmml.to_pmml(model_output, pmml_output, model_type="sklearn")
self.assertTrue(os.path.exists(pmml_output))
# validate pmml file
data_output = os.path.join(self.data.base_path,
"{}.csv".format(algo_name))
ppmml.predict(pmml_output, self.data.test_data_input, data_output)
self.assertTrue(os.path.exists(data_output))
def test_exceptions(self):
""" test exceptions
"""
with self.assertRaises(ValueError):
ppmml.to_pmml("non_existent_path", "./pmml_output.pmml")