tf.io / match_filenames_once

TensorFlow 1 version

View source on GitHub

Parses a batch of SequenceExample protos.

tf.io.parse_sequence_example(
    serialized, context_features=None, sequence_features=None, example_names=None,
    name=None
)

Parses a vector of serialized SequenceExample protos given in serialized.

This op parses serialized sequence examples into a tuple of dictionaries, each mapping keys to Tensor and SparseTensor objects. The first dictionary contains mappings for keys appearing in context_features, and the second dictionary contains mappings for keys appearing in sequence_features.

At least one of context_features and sequence_features must be provided and non-empty.

The context_features keys are associated with a SequenceExample as a whole, independent of time / frame. In contrast, the sequence_features keys provide a way to access variable-length data within the FeatureList section of the SequenceExample proto. While the shapes of context_features values are fixed with respect to frame, the frame dimension (the first dimension) of sequence_features values may vary between SequenceExample protos, and even between feature_list keys within the same SequenceExample.

context_features contains VarLenFeature, RaggedFeature, and FixedLenFeature objects. Each VarLenFeature is mapped to a SparseTensor; each RaggedFeature is mapped to a RaggedTensor; and each FixedLenFeature is mapped to a Tensor, of the specified type, shape, and default value.

sequence_features contains VarLenFeature, RaggedFeature, and FixedLenSequenceFeature objects. Each VarLenFeature is mapped to a SparseTensor; each RaggedFeature is mapped to a RaggedTensor; and eachFixedLenSequenceFeatureis mapped to aTensor, each of the specified type. The shape will be(B,T,) + df.dense_shapeforFixedLenSequenceFeaturedf, whereBis the batch size, andTis the length of the associatedFeatureListin theSequenceExample. For instance,FixedLenSequenceFeature([])yields a scalar 2-DTensorof static shape[None, None]and dynamic shape[B, T], whileFixedLenSequenceFeature([k])(forint k >= 1) yields a 3-D matrixTensorof static shape[None, None, k]and dynamic shape[B, T, k]`.

Like the input, the resulting output tensors have a batch dimension. This means that the original per-example shapes of VarLenFeatures and FixedLenSequenceFeatures can be lost. To handle that situation, this op also provides dicts of shape tensors as part of the output. There is one dict for the context features, and one for the feature_list features. Context features of type FixedLenFeatures will not be present, since their shapes are already known by the caller. In situations where the input 'FixedLenFeature`s are of different lengths across examples, the shorter examples will be padded with default datatype values: 0 for numeric types, and the empty string for string types.

Each SparseTensor corresponding to sequence_features represents a ragged vector. Its indices are [time, index], where time is the FeatureList entry and index is the value's index in the list of values associated with that time.

FixedLenFeature entries with a default_value and FixedLenSequenceFeature entries with allow_missing=True are optional; otherwise, we will fail if that Feature or FeatureList is missing from any example in serialized.

example_name may contain a descriptive name for the corresponding serialized proto. This may be useful for debugging purposes, but it has no effect on the output. If not None, example_name must be a scalar.