import copy
import functools
import warnings
from types import MethodType
from typing import Dict, List, Optional, Type, Union
import dill
import pandas as pd
from feast.base_feature_view import BaseFeatureView
from feast.data_source import RequestSource
from feast.errors import RegistryInferenceFailure, SpecifiedFeaturesNotPresentError
from feast.feature import Feature
from feast.feature_view import FeatureView
from feast.feature_view_projection import FeatureViewProjection
from feast.field import Field, from_value_type
from feast.protos.feast.core.OnDemandFeatureView_pb2 import (
OnDemandFeatureView as OnDemandFeatureViewProto,
)
from feast.protos.feast.core.OnDemandFeatureView_pb2 import (
OnDemandFeatureViewMeta,
OnDemandFeatureViewSpec,
OnDemandSource,
)
from feast.protos.feast.core.OnDemandFeatureView_pb2 import (
UserDefinedFunction as UserDefinedFunctionProto,
)
from feast.type_map import (
feast_value_type_to_pandas_type,
python_type_to_feast_value_type,
)
from feast.usage import log_exceptions
from feast.value_type import ValueType
warnings.simplefilter("once", DeprecationWarning)
[docs]class OnDemandFeatureView(BaseFeatureView):
"""
[Experimental] An OnDemandFeatureView defines a logical group of features that are
generated by applying a transformation on a set of input sources, such as feature
views and request data sources.
Attributes:
name: The unique name of the on demand feature view.
features: The list of features in the output of the on demand feature view.
source_feature_view_projections: A map from input source names to actual input
sources with type FeatureViewProjection.
source_request_sources: A map from input source names to the actual input
sources with type RequestSource.
udf: The user defined transformation function, which must take pandas dataframes
as inputs.
description: A human-readable description.
tags: A dictionary of key-value pairs to store arbitrary metadata.
owner: The owner of the on demand feature view, typically the email of the primary
maintainer.
"""
# TODO(adchia): remove inputs from proto and declaration
name: str
features: List[Field]
source_feature_view_projections: Dict[str, FeatureViewProjection]
source_request_sources: Dict[str, RequestSource]
udf: MethodType
description: str
tags: Dict[str, str]
owner: str
@log_exceptions
def __init__(
self,
*args,
name: Optional[str] = None,
features: Optional[List[Feature]] = None,
sources: Optional[
Dict[str, Union[FeatureView, FeatureViewProjection, RequestSource]]
] = None,
udf: Optional[MethodType] = None,
inputs: Optional[
Dict[str, Union[FeatureView, FeatureViewProjection, RequestSource]]
] = None,
schema: Optional[List[Field]] = None,
description: str = "",
tags: Optional[Dict[str, str]] = None,
owner: str = "",
):
"""
Creates an OnDemandFeatureView object.
Args:
name: The unique name of the on demand feature view.
features (deprecated): The list of features in the output of the on demand
feature view, after the transformation has been applied.
sources (optional): A map from input source names to the actual input sources,
which may be feature views, feature view projections, or request data sources.
These sources serve as inputs to the udf, which will refer to them by name.
udf (optional): The user defined transformation function, which must take pandas
dataframes as inputs.
inputs (optional): A map from input source names to the actual input sources,
which may be feature views, feature view projections, or request data sources.
These sources serve as inputs to the udf, which will refer to them by name.
schema (optional): The list of features in the output of the on demand feature
view, after the transformation has been applied.
description (optional): A human-readable description.
tags (optional): A dictionary of key-value pairs to store arbitrary metadata.
owner (optional): The owner of the on demand feature view, typically the email
of the primary maintainer.
"""
positional_attributes = ["name", "features", "inputs", "udf"]
_name = name
_schema = schema or []
if len(_schema) == 0 and features is not None:
_schema = [Field.from_feature(feature) for feature in features]
if features is not None:
warnings.warn(
(
"The `features` parameter is being deprecated in favor of the `schema` parameter. "
"Please switch from using `features` to `schema`. This will also requiring switching "
"feature definitions from using `Feature` to `Field`. Feast 0.21 and onwards will not "
"support the `features` parameter."
),
DeprecationWarning,
)
_sources = sources or inputs
if inputs and sources:
raise ValueError("At most one of `sources` or `inputs` can be specified.")
elif inputs:
warnings.warn(
(
"The `inputs` parameter is being deprecated. Please use `sources` instead. "
"Feast 0.21 and onwards will not support the `inputs` parameter."
),
DeprecationWarning,
)
_udf = udf
if args:
warnings.warn(
(
"On demand feature view parameters should be specified as keyword arguments "
"instead of positional arguments. Feast 0.23 and onwards will not support "
"positional arguments in on demand feature view definitions."
),
DeprecationWarning,
)
if len(args) > len(positional_attributes):
raise ValueError(
f"Only {', '.join(positional_attributes)} are allowed as positional args "
f"when defining feature views, for backwards compatibility."
)
if len(args) >= 1:
_name = args[0]
if len(args) >= 2:
_schema = args[1]
# Convert Features to Fields.
if len(_schema) > 0 and isinstance(_schema[0], Feature):
_schema = [Field.from_feature(feature) for feature in _schema]
warnings.warn(
(
"The `features` parameter is being deprecated in favor of the `schema` parameter. "
"Please switch from using `features` to `schema`. This will also requiring switching "
"feature definitions from using `Feature` to `Field`. Feast 0.21 and onwards will not "
"support the `features` parameter."
),
DeprecationWarning,
)
if len(args) >= 3:
_sources = args[2]
warnings.warn(
(
"The `inputs` parameter is being deprecated. Please use `sources` instead. "
"Feast 0.21 and onwards will not support the `inputs` parameter."
),
DeprecationWarning,
)
if len(args) >= 4:
_udf = args[3]
if not _name:
raise ValueError(
"The name of the on demand feature view must be specified."
)
if not _sources:
raise ValueError("The `sources` parameter must be specified.")
super().__init__(
name=_name,
features=_schema,
description=description,
tags=tags,
owner=owner,
)
assert _sources is not None
self.source_feature_view_projections: Dict[str, FeatureViewProjection] = {}
self.source_request_sources: Dict[str, RequestSource] = {}
for source_name, odfv_source in _sources.items():
if isinstance(odfv_source, RequestSource):
self.source_request_sources[source_name] = odfv_source
elif isinstance(odfv_source, FeatureViewProjection):
self.source_feature_view_projections[source_name] = odfv_source
else:
self.source_feature_view_projections[
source_name
] = odfv_source.projection
if _udf is None:
raise ValueError("The `udf` parameter must be specified.")
assert _udf
self.udf = _udf
@property
def proto_class(self) -> Type[OnDemandFeatureViewProto]:
return OnDemandFeatureViewProto
def __copy__(self):
fv = OnDemandFeatureView(
name=self.name,
schema=self.features,
sources=dict(
**self.source_feature_view_projections, **self.source_request_sources,
),
udf=self.udf,
description=self.description,
tags=self.tags,
owner=self.owner,
)
fv.projection = copy.copy(self.projection)
return fv
def __eq__(self, other):
if not isinstance(other, OnDemandFeatureView):
raise TypeError(
"Comparisons should only involve OnDemandFeatureView class objects."
)
if not super().__eq__(other):
return False
if (
self.source_feature_view_projections
!= other.source_feature_view_projections
or self.source_request_sources != other.source_request_sources
or self.udf.__code__.co_code != other.udf.__code__.co_code
):
return False
return True
def __hash__(self):
return super().__hash__()
[docs] def to_proto(self) -> OnDemandFeatureViewProto:
"""
Converts an on demand feature view object to its protobuf representation.
Returns:
A OnDemandFeatureViewProto protobuf.
"""
meta = OnDemandFeatureViewMeta()
if self.created_timestamp:
meta.created_timestamp.FromDatetime(self.created_timestamp)
if self.last_updated_timestamp:
meta.last_updated_timestamp.FromDatetime(self.last_updated_timestamp)
sources = {}
for source_name, fv_projection in self.source_feature_view_projections.items():
sources[source_name] = OnDemandSource(
feature_view_projection=fv_projection.to_proto()
)
for (source_name, request_sources,) in self.source_request_sources.items():
sources[source_name] = OnDemandSource(
request_data_source=request_sources.to_proto()
)
spec = OnDemandFeatureViewSpec(
name=self.name,
features=[feature.to_proto() for feature in self.features],
sources=sources,
user_defined_function=UserDefinedFunctionProto(
name=self.udf.__name__, body=dill.dumps(self.udf, recurse=True),
),
description=self.description,
tags=self.tags,
owner=self.owner,
)
return OnDemandFeatureViewProto(spec=spec, meta=meta)
[docs] @classmethod
def from_proto(cls, on_demand_feature_view_proto: OnDemandFeatureViewProto):
"""
Creates an on demand feature view from a protobuf representation.
Args:
on_demand_feature_view_proto: A protobuf representation of an on-demand feature view.
Returns:
A OnDemandFeatureView object based on the on-demand feature view protobuf.
"""
sources = {}
for (
source_name,
on_demand_source,
) in on_demand_feature_view_proto.spec.sources.items():
if on_demand_source.WhichOneof("source") == "feature_view":
sources[source_name] = FeatureView.from_proto(
on_demand_source.feature_view
).projection
elif on_demand_source.WhichOneof("source") == "feature_view_projection":
sources[source_name] = FeatureViewProjection.from_proto(
on_demand_source.feature_view_projection
)
else:
sources[source_name] = RequestSource.from_proto(
on_demand_source.request_data_source
)
on_demand_feature_view_obj = cls(
name=on_demand_feature_view_proto.spec.name,
schema=[
Field(
name=feature.name,
dtype=from_value_type(ValueType(feature.value_type)),
)
for feature in on_demand_feature_view_proto.spec.features
],
sources=sources,
udf=dill.loads(
on_demand_feature_view_proto.spec.user_defined_function.body
),
description=on_demand_feature_view_proto.spec.description,
tags=dict(on_demand_feature_view_proto.spec.tags),
owner=on_demand_feature_view_proto.spec.owner,
)
# FeatureViewProjections are not saved in the OnDemandFeatureView proto.
# Create the default projection.
on_demand_feature_view_obj.projection = FeatureViewProjection.from_definition(
on_demand_feature_view_obj
)
if on_demand_feature_view_proto.meta.HasField("created_timestamp"):
on_demand_feature_view_obj.created_timestamp = (
on_demand_feature_view_proto.meta.created_timestamp.ToDatetime()
)
if on_demand_feature_view_proto.meta.HasField("last_updated_timestamp"):
on_demand_feature_view_obj.last_updated_timestamp = (
on_demand_feature_view_proto.meta.last_updated_timestamp.ToDatetime()
)
return on_demand_feature_view_obj
[docs] def get_request_data_schema(self) -> Dict[str, ValueType]:
schema: Dict[str, ValueType] = {}
for request_source in self.source_request_sources.values():
if isinstance(request_source.schema, List):
new_schema = {}
for field in request_source.schema:
new_schema[field.name] = field.dtype.to_value_type()
schema.update(new_schema)
elif isinstance(request_source.schema, Dict):
schema.update(request_source.schema)
else:
raise Exception(
f"Request source schema is not correct type: ${str(type(request_source.schema))}"
)
return schema
[docs] def infer_features(self):
"""
Infers the set of features associated to this feature view from the input source.
Raises:
RegistryInferenceFailure: The set of features could not be inferred.
"""
df = pd.DataFrame()
for feature_view_projection in self.source_feature_view_projections.values():
for feature in feature_view_projection.features:
dtype = feast_value_type_to_pandas_type(feature.dtype.to_value_type())
df[f"{feature_view_projection.name}__{feature.name}"] = pd.Series(
dtype=dtype
)
df[f"{feature.name}"] = pd.Series(dtype=dtype)
for request_data in self.source_request_sources.values():
for field in request_data.schema:
dtype = feast_value_type_to_pandas_type(field.dtype.to_value_type())
df[f"{field.name}"] = pd.Series(dtype=dtype)
output_df: pd.DataFrame = self.udf.__call__(df)
inferred_features = []
for f, dt in zip(output_df.columns, output_df.dtypes):
inferred_features.append(
Field(
name=f,
dtype=from_value_type(
python_type_to_feast_value_type(f, type_name=str(dt))
),
)
)
if self.features:
missing_features = []
for specified_features in self.features:
if specified_features not in inferred_features:
missing_features.append(specified_features)
if missing_features:
raise SpecifiedFeaturesNotPresentError(
[f.name for f in missing_features], self.name
)
else:
self.features = inferred_features
if not self.features:
raise RegistryInferenceFailure(
"OnDemandFeatureView",
f"Could not infer Features for the feature view '{self.name}'.",
)
[docs] @staticmethod
def get_requested_odfvs(feature_refs, project, registry):
all_on_demand_feature_views = registry.list_on_demand_feature_views(
project, allow_cache=True
)
requested_on_demand_feature_views: List[OnDemandFeatureView] = []
for odfv in all_on_demand_feature_views:
for feature in odfv.features:
if f"{odfv.name}:{feature.name}" in feature_refs:
requested_on_demand_feature_views.append(odfv)
break
return requested_on_demand_feature_views
# TODO(felixwang9817): Force this decorator to accept kwargs and switch from
# `features` to `schema`.
[docs]def on_demand_feature_view(
*args,
features: Optional[List[Feature]] = None,
sources: Optional[Dict[str, Union[FeatureView, RequestSource]]] = None,
inputs: Optional[Dict[str, Union[FeatureView, RequestSource]]] = None,
schema: Optional[List[Field]] = None,
description: str = "",
tags: Optional[Dict[str, str]] = None,
owner: str = "",
):
"""
Creates an OnDemandFeatureView object with the given user function as udf.
Args:
features (deprecated): The list of features in the output of the on demand
feature view, after the transformation has been applied.
sources (optional): A map from input source names to the actual input sources,
which may be feature views, feature view projections, or request data sources.
These sources serve as inputs to the udf, which will refer to them by name.
inputs (optional): A map from input source names to the actual input sources,
which may be feature views, feature view projections, or request data sources.
These sources serve as inputs to the udf, which will refer to them by name.
schema (optional): The list of features in the output of the on demand feature
view, after the transformation has been applied.
description (optional): A human-readable description.
tags (optional): A dictionary of key-value pairs to store arbitrary metadata.
owner (optional): The owner of the on demand feature view, typically the email
of the primary maintainer.
"""
positional_attributes = ["features", "inputs"]
_schema = schema or []
if len(_schema) == 0 and features is not None:
_schema = [Field.from_feature(feature) for feature in features]
if features is not None:
warnings.warn(
(
"The `features` parameter is being deprecated in favor of the `schema` parameter. "
"Please switch from using `features` to `schema`. This will also requiring switching "
"feature definitions from using `Feature` to `Field`. Feast 0.21 and onwards will not "
"support the `features` parameter."
),
DeprecationWarning,
)
_sources = sources or inputs
if inputs and sources:
raise ValueError("At most one of `sources` or `inputs` can be specified.")
elif inputs:
warnings.warn(
(
"The `inputs` parameter is being deprecated. Please use `sources` instead. "
"Feast 0.21 and onwards will not support the `inputs` parameter."
),
DeprecationWarning,
)
if args:
warnings.warn(
(
"On demand feature view parameters should be specified as keyword arguments "
"instead of positional arguments. Feast 0.23 and onwards will not support "
"positional arguments in on demand feature view definitions."
),
DeprecationWarning,
)
if len(args) > len(positional_attributes):
raise ValueError(
f"Only {', '.join(positional_attributes)} are allowed as positional args "
f"when defining feature views, for backwards compatibility."
)
if len(args) >= 1:
_schema = args[0]
# Convert Features to Fields.
if len(_schema) > 0 and isinstance(_schema[0], Feature):
_schema = [Field.from_feature(feature) for feature in _schema]
warnings.warn(
(
"The `features` parameter is being deprecated in favor of the `schema` parameter. "
"Please switch from using `features` to `schema`. This will also requiring switching "
"feature definitions from using `Feature` to `Field`. Feast 0.21 and onwards will not "
"support the `features` parameter."
),
DeprecationWarning,
)
if len(args) >= 2:
_sources = args[1]
warnings.warn(
(
"The `inputs` parameter is being deprecated. Please use `sources` instead. "
"Feast 0.21 and onwards will not support the `inputs` parameter."
),
DeprecationWarning,
)
if not _sources:
raise ValueError("The `sources` parameter must be specified.")
def decorator(user_function):
on_demand_feature_view_obj = OnDemandFeatureView(
name=user_function.__name__,
sources=_sources,
schema=_schema,
udf=user_function,
description=description,
tags=tags,
owner=owner,
)
functools.update_wrapper(
wrapper=on_demand_feature_view_obj, wrapped=user_function
)
return on_demand_feature_view_obj
return decorator