A TensorAdapter converts a RecordBatch to a collection of TF Tensors.
The conversion is determined by both the Arrow schema and the TensorRepresentations, which must be provided at the initialization time. Each TensorRepresentation contains the information needed to translates one or more columns in a RecordBatch of the given Arrow schema into a TF Tensor or CompositeTensor. They are contained in a Dict whose keys are the names of the tensors, which will be the keys of the Dict produced by ToBatchTensors().
TypeSpecs() returns static TypeSpecs of those tensors by their names, i.e. if they have a shape, then the size of the first (batch) dimension is always unknown (None) because it depends on the size of the RecordBatch passed to ToBatchTensors().
It is guaranteed that for any tensor_name in the given TensorRepresentations self.TypeSpecs()[tensor_name].is_compatible_with( self.ToBatchedTensors(...)[tensor_name])
OriginalTypeSpecs() -> Dict[Text, common_types.TensorTypeSpec]
Returns the origin's type specs.
A TFXIO 'Y' may be a result of projection of another TFXIO 'X', in which case then 'X' is the origin of 'Y'. And this method returns what X.TensorAdapter().TypeSpecs() would return.
May equal to
Returns: a mapping from tensor names to
ToBatchTensors( record_batch: pa.RecordBatch, produce_eager_tensors: Optional[bool] = None ) -> Dict[Text, Any]
Returns a batch of tensors translated from
||controls whether the ToBatchTensors() produces eager tensors or ndarrays (or Tensor value objects). If None, determine that from whether TF Eager mode is enabled.|
||when Eager Tensors are requested but TF is not executing eagerly.|
||when Any handler failed to produce a Tensor.|
TypeSpecs() -> Dict[Text, common_types.TensorTypeSpec]
Returns the TypeSpec for each tensor.