def _get_loss()

in tensorflow_model_remediation/min_diff/losses/loss_utils.py [0:0]

• 16 lines of code
• 5 McCabe index (conditional complexity)

def _get_loss(loss: Union[base_loss.MinDiffLoss, Text],
              loss_var_name: Text = 'loss') -> base_loss.MinDiffLoss:
  """Returns a `losses.MinDiffLoss` instance corresponding to `loss`.

  If `loss` is an instance of `losses.MinDiffLoss` then it is returned
  directly. If `loss` is a string it must be an accepted loss name. A
  value of `None` is also accepted and simply returns `None`.

  Args:
    loss: loss instance. Can be `None`, a string or an instance of
      `losses.MinDiffLoss`.
    loss_var_name: Name of the loss variable. This is only used for error
      messaging.

  Returns:
    A `MinDiffLoss` instance.

  Raises:
    ValueError: If `loss` is an unrecognized string.
    TypeError: If `loss` is not an instance of `losses.MinDiffLoss` or a string.
  """
  if loss is None:
    return None
  if isinstance(loss, base_loss.MinDiffLoss):
    return loss
  if isinstance(loss, six.string_types):
    lower_case_loss = loss.lower()
    if lower_case_loss in _STRING_TO_LOSS_DICT:
      return _STRING_TO_LOSS_DICT[lower_case_loss]()
    raise ValueError('If {} is a string, it must be a (case-insensitive) '
                     'match for one of the following supported values: {}. '
                     'given: {}'.format(loss_var_name,
                                        _STRING_TO_LOSS_DICT.keys(), loss))
  raise TypeError('{} must be either of type MinDiffLoss or string, given: '
                  '{} (type: {})'.format(loss_var_name, loss, type(loss)))