Function locate_tensor_element

tensorflow/python/debug/cli/tensor_format.py:282–403

Locate a tensor element in formatted text lines, given element indices. Given a RichTextLines object representing a tensor and indices of the sought element, return the row number at which the element is located (if exists). Args: formatted: A RichTextLines object containing formatted te

(formatted, indices)

def locate_tensor_element(formatted, indices):
  """Locate a tensor element in formatted text lines, given element indices.

  Given a RichTextLines object representing a tensor and indices of the sought
  element, return the row number at which the element is located (if exists).

  Args:
    formatted: A RichTextLines object containing formatted text lines
      representing the tensor.
    indices: Indices of the sought element, as a list of int or a list of list
      of int. The former case is for a single set of indices to look up,
      whereas the latter case is for looking up a batch of indices sets at once.
      In the latter case, the indices must be in ascending order, or a
      ValueError will be raised.

  Returns:
    1) A boolean indicating whether the element falls into an omitted line.
    2) Row index.
    3) Column start index, i.e., the first column in which the representation
       of the specified tensor starts, if it can be determined. If it cannot
       be determined (e.g., due to ellipsis), None.
    4) Column end index, i.e., the column right after the last column that
       represents the specified tensor. Iff it cannot be determined, None.

  For return values described above are based on a single set of indices to
    look up. In the case of batch mode (multiple sets of indices), the return
    values will be lists of the types described above.

  Raises:
    AttributeError: If:
      Input argument "formatted" does not have the required annotations.
    ValueError: If:
      1) Indices do not match the dimensions of the tensor, or
      2) Indices exceed sizes of the tensor, or
      3) Indices contain negative value(s).
      4) If in batch mode, and if not all sets of indices are in ascending
         order.
  """

  if isinstance(indices[0], list):
    indices_list = indices
    input_batch = True
  else:
    indices_list = [indices]
    input_batch = False

  # Check that tensor_metadata is available.
  if "tensor_metadata" not in formatted.annotations:
    raise AttributeError("tensor_metadata is not available in annotations.")

  # Sanity check on input argument.
  _validate_indices_list(indices_list, formatted)

  dims = formatted.annotations["tensor_metadata"]["shape"]
  batch_size = len(indices_list)
  lines = formatted.lines
  annot = formatted.annotations
  prev_r = 0