tensorflow_privacy/privacy/estimators/binary_class_head.py [40:86]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        loss_fn=loss_fn,
        name=name)

  def loss(self,
           labels,
           logits,
           features=None,
           mode=None,
           regularization_losses=None):
    """Returns regularized training loss. See `base_head.Head` for details."""
    del mode  # Unused for this head.
    with tf.compat.v1.name_scope(
        'losses', values=(logits, labels, regularization_losses, features)):
      logits = base_head.check_logits_final_dim(logits, self.logits_dimension)
      labels = self._processed_labels(logits, labels)
      unweighted_loss, weights = self._unweighted_loss_and_weights(
          logits, labels, features)
      vector_training_loss = losses_utils.compute_weighted_loss(
          unweighted_loss,
          sample_weight=weights,
          reduction=tf.keras.losses.Reduction.NONE)
      regularization_loss = tf.math.add_n(
          regularization_losses) if regularization_losses is not None else None
      vector_regularized_training_loss = (
          tf.add(vector_training_loss, regularization_loss)
          if regularization_loss is not None else vector_training_loss)

    return vector_regularized_training_loss

  def _create_tpu_estimator_spec(self,
                                 features,
                                 mode,
                                 logits,
                                 labels=None,
                                 optimizer=None,
                                 trainable_variables=None,
                                 train_op_fn=None,
                                 update_ops=None,
                                 regularization_losses=None):
    """See superclass for description."""

    with tf.compat.v1.name_scope(self._name, 'head'):
      # Predict.
      pred_keys = prediction_keys.PredictionKeys
      predictions = self.predictions(logits)
      if mode == ModeKeys.PREDICT:
        probabilities = predictions[pred_keys.PROBABILITIES]
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



tensorflow_privacy/privacy/estimators/multi_class_head.py [40:86]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        loss_fn=loss_fn,
        name=name)

  def loss(self,
           labels,
           logits,
           features=None,
           mode=None,
           regularization_losses=None):
    """Returns regularized training loss. See `base_head.Head` for details."""
    del mode  # Unused for this head.
    with tf.compat.v1.name_scope(
        'losses', values=(logits, labels, regularization_losses, features)):
      logits = base_head.check_logits_final_dim(logits, self.logits_dimension)
      labels = self._processed_labels(logits, labels)
      unweighted_loss, weights = self._unweighted_loss_and_weights(
          logits, labels, features)
      vector_training_loss = losses_utils.compute_weighted_loss(
          unweighted_loss,
          sample_weight=weights,
          reduction=tf.keras.losses.Reduction.NONE)
      regularization_loss = tf.math.add_n(
          regularization_losses) if regularization_losses is not None else None
      vector_regularized_training_loss = (
          tf.add(vector_training_loss, regularization_loss)
          if regularization_loss is not None else vector_training_loss)

    return vector_regularized_training_loss

  def _create_tpu_estimator_spec(self,
                                 features,
                                 mode,
                                 logits,
                                 labels=None,
                                 optimizer=None,
                                 trainable_variables=None,
                                 train_op_fn=None,
                                 update_ops=None,
                                 regularization_losses=None):
    """See superclass for description."""

    with tf.compat.v1.name_scope(self._name, 'head'):
      # Predict.
      pred_keys = prediction_keys.PredictionKeys
      predictions = self.predictions(logits)
      if mode == ModeKeys.PREDICT:
        probabilities = predictions[pred_keys.PROBABILITIES]
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -