def parse_schema_training_dataset(self, dataframe):
return [
training_dataset_feature.TrainingDatasetFeature(
feat.name.lower(), feat.dataType.simpleString()
)
for feat in dataframe.schema
]
def attach_transformation_fn(
self, training_dataset_obj=None, feature_view_obj=None
):
if training_dataset_obj:
target_obj = training_dataset_obj
else:
target_obj = feature_view_obj
if target_obj._transformation_functions:
for (
feature_name,
transformation_fn,
) in target_obj._transformation_functions.items():
if feature_name in target_obj.labels:
raise ValueError(
"Online transformations for training dataset labels are not supported."
)
target_obj._features.append(
training_dataset_feature.TrainingDatasetFeature(
name=feature_name,
feature_group_feature_name=feature_name,
type=transformation_fn.output_type,
label=False,
transformation_function=transformation_fn,
)
)
def save(self, feature_view_obj):
if feature_view_obj.labels:
feature_view_obj._features.append([
training_dataset_feature.TrainingDatasetFeature(
name=label_name, label=True)
for label_name in feature_view_obj.labels
])
self._transformation_function_engine.attach_transformation_fn(
feature_view_obj)
updated_fv = self._feature_view_api.post(feature_view_obj)
print("Feature view created successfully, explore it at \n" +
self._get_feature_view_url(updated_fv))
return updated_fv
def save(self, training_dataset, features, user_write_options):
if isinstance(features, query.Query):
training_dataset._querydto = features
training_dataset._features = [
training_dataset_feature.TrainingDatasetFeature(
name=label_name, label=True)
for label_name in training_dataset.label
]
self._transformation_function_engine.attach_transformation_fn(
training_dataset)
else:
features = engine.get_instance().convert_to_default_dataframe(
features)
training_dataset._features = (
engine.get_instance().parse_schema_training_dataset(features))
for label_name in training_dataset.label:
for feature in training_dataset._features:
if feature.name == label_name:
feature.label = True
# check if user provided transformation functions and throw error as transformation functions work only
# with query objects
if training_dataset.transformation_functions:
raise ValueError(
"Transformation functions can only be applied to training datasets generated from Query object"
)
if len(training_dataset.splits
) > 0 and training_dataset.train_split is None:
training_dataset.train_split = "train"
warnings.warn(
"Training dataset splits were defined but no `train_split` (the name of the split that is going to be "
"used for training) was provided. Setting this property to `train`. The statistics of this "
"split will be used for transformation functions.")
updated_instance = self._training_dataset_api.post(training_dataset)
td_job = engine.get_instance().write_training_dataset(
training_dataset, features, user_write_options, self.OVERWRITE)
return updated_instance, td_job
def save(self, training_dataset, features, user_write_options):
if isinstance(features, query.Query):
training_dataset._querydto = features
training_dataset._features = [
training_dataset_feature.TrainingDatasetFeature(
name=label_name, label=True
)
for label_name in training_dataset.label
]
else:
features = engine.get_instance().convert_to_default_dataframe(features)
training_dataset._features = (
engine.get_instance().parse_schema_training_dataset(features)
)
for label_name in training_dataset.label:
for feature in training_dataset._features:
if feature.name == label_name:
feature.label = True
self._training_dataset_api.post(training_dataset)
engine.get_instance().write_training_dataset(
training_dataset, features, user_write_options, self.OVERWRITE
)