def extract_features(
self,
features: features_lib.FeatureConnector,
) -> features_lib.FeatureConnector:
"""Returns the `tfds.features.FeaturesDict`.
Extract the subset of features
Args:
features: Features on which extract the sub-set
Returns:
features_subset: A subset of the features
"""
with utils.try_reraise(
'Provided PartialDecoding specs does not match actual features: '
):
# Convert non-features into features
expected_feature = _normalize_feature_item(
feature=features,
expected_feature=self._feature_specs,
)
# Get the intersection of `features` and `expected_feature`
return _extract_features(
feature=features,
expected_feature=features_dict.to_feature(expected_feature),
)
def __init__(self, feature, length=None, **kwargs):
"""Construct a sequence dict.
Args:
feature: `dict`, the features to wrap
length: `int`, length of the sequence if static and known in advance
**kwargs: `dict`, constructor kwargs of `tfds.features.FeaturesDict`
"""
# Convert {} => FeaturesDict, tf.int32 => Tensor(shape=(), dtype=tf.int32)
self._feature = features_dict.to_feature(feature)
self._length = length
super(Sequence, self).__init__(**kwargs)
def __init__(
self,
feature: feature_lib.FeatureConnectorArg,
length: Optional[int] = None,
):
"""Construct a sequence dict.
Args:
feature: The features to wrap (any feature supported)
length: `int`, length of the sequence if static and known in advance
"""
# Convert {} => FeaturesDict, tf.int32 => Tensor(shape=(), dtype=tf.int32)
self._feature = features_dict.to_feature(feature)
self._length = length
super(Sequence, self).__init__()
