def _feature_from_feature_spec(spec, name, domains):
"""Returns a representation of a Feature from a feature spec."""
if isinstance(spec, tf.io.FixedLenFeature):
if spec.default_value is not None:
raise ValueError(
'feature "{}" had default_value {}, but FixedLenFeature must have '
'default_value=None'.format(name, spec.default_value))
dims = [schema_pb2.FixedShape.Dim(size=size) for size in spec.shape]
feature = schema_pb2.Feature(
name=name,
presence=schema_pb2.FeaturePresence(min_fraction=1.0),
shape=schema_pb2.FixedShape(dim=dims))
elif isinstance(spec, tf.io.VarLenFeature):
feature = schema_pb2.Feature(name=name)
else:
raise TypeError(
'Spec for feature "{}" was {} of type {}, expected a '
'FixedLenFeature, VarLenFeature or SparseFeature'.format(
name, spec, type(spec)))
_set_type(name, feature, spec.dtype)
_set_domain(name, feature, domains.get(name))
return feature
def _legacy_schema_from_feature_spec(feature_spec, domains=None):
"""Infer a Schema from a feature spec, using the legacy feature spec logic.
Infers a Schema proto that with generate_legacy_feature_spec set to true,
which will result in the given feature spec and domains when
schema_as_feature_spec is called. This is used to represent feature specs
that can only be represented when generate_legacy_feature_spec is true. In
particular, feature specs with a default value set.
Args:
feature_spec: A TensorFlow feature spec
domains: A dict from key names to `IntDomain`s
Returns:
A Schema proto.
Raises:
ValueError: If a default value is invalid.
TypeError: If an unknown type of feature spec is encountered.
"""
result = schema_pb2.Schema()
result.generate_legacy_feature_spec = True
for name, spec in sorted(feature_spec.items()):
if isinstance(spec, tf.io.FixedLenFeature):
# Validate shape first as shape governs which default values are valid.
if len(spec.shape) == 0: # pylint: disable=g-explicit-length-test
size = 1
expected_default_value = '' if spec.dtype == tf.string else -1
elif len(spec.shape) == 1 and spec.shape[0] > 1:
size = spec.shape[0]
expected_default_value = [
'' if spec.dtype == tf.string else -1
] * size
else:
raise ValueError(
'When inferring legacy schema from feature spec, feature "{}" had '
'shape {}, but FixedLenFeature must have shape [] or [k] where '
'k > 1.'.format(name, spec.shape))
if spec.default_value is None:
min_fraction = 1
elif spec.default_value == expected_default_value:
min_fraction = 0
else:
raise ValueError(
'When inferring legacy schema from feature spec, feature "{}" had '
'default_value {}, but FixedLenFeature must have '
'default_value=None or {}'.format(name, spec.default_value,
expected_default_value))
feature = result.feature.add(
name=name,
presence=schema_pb2.FeaturePresence(min_fraction=min_fraction),
value_count=schema_pb2.ValueCount(min=size, max=size))
elif isinstance(spec, tf.io.VarLenFeature):
feature = result.feature.add(name=name)
else:
raise TypeError(
'When inferring legacy schema from feature spec, spec for feature '
'"{}" was {} of type {}, expected a FixedLenFeature or '
'VarLenFeature '.format(name, spec, type(spec)))
_set_type(name, feature, spec.dtype)
_set_domain(name, feature, domains.get(name))
return result