Data

Data can be emitted by Datasets (i.e., the element types of all Datasets are Data).

Currently supported data types are:

• Single Tensor.
• Sequences of other Data (e.g., Seqs, Lists, etc.).
  • Sequences that are not homogeneous are not supported (e.g., Seq(data1, Seq(data1, data2))).
  • Note that, for that reason, even though Seq(List(data1), List(data1, data2)) is supported, Seq(Seq(data1), List(data1, data2)) is not.
  • A sequence containing both Outputs and SparseOutputs, for example, is considered heterogeneous. For such cases, it is advisable to use tuples.
• Arrays of other Data.
• Maps with arbitrary key types and Data value types.
• Products of other Data (e.g., tuples).
  • Note that with tuples, heterogeneous types are supported, due to the tuple type being a kind of heterogeneous collection. Internally, the data emitted by a Dataset will be de-duplicated to prevent redundant computation.

This trait guarantees that the output data types and shapes of a Dataset will match the structure of the corresponding data. For example, if a Seq(List(data1), List(data1, data2)) is provided as a Dataset element type, then the dataset output data types will have the following structure Seq(List(type1), List(type1, type2)), and similarly for the output shapes.