Python tensorflow.python.ops.math_ops.to_float() Examples

def create_test_input(batch_size, height, width, channels):
  """Create test input tensor.

  Args:
    batch_size: The number of images per batch or `None` if unknown.
    height: The height of each image or `None` if unknown.
    width: The width of each image or `None` if unknown.
    channels: The number of channels per image or `None` if unknown.

  Returns:
    Either a placeholder `Tensor` of dimension
      [batch_size, height, width, channels] if any of the inputs are `None` or a
    constant `Tensor` with the mesh grid values along the spatial dimensions.
  """
  if None in [batch_size, height, width, channels]:
    return array_ops.placeholder(dtypes.float32,
                                 (batch_size, height, width, channels))
  else:
    return math_ops.to_float(
        np.tile(
            np.reshape(
                np.reshape(np.arange(height), [height, 1]) + np.reshape(
                    np.arange(width), [1, width]), [1, height, width, 1]),
            [batch_size, 1, 1, channels])) 

def _lengths_to_masks(lengths, max_length):
  """Creates a binary matrix that can be used to mask away padding.

  Args:
    lengths: A vector of integers representing lengths.
    max_length: An integer indicating the maximum length. All values in
      lengths should be less than max_length.
  Returns:
    masks: Masks that can be used to get rid of padding.
  """
  tiled_ranges = array_ops.tile(
      array_ops.expand_dims(math_ops.range(max_length), 0),
      [array_ops.shape(lengths)[0], 1])
  lengths = array_ops.expand_dims(lengths, 1)
  masks = math_ops.to_float(
      math_ops.to_int64(tiled_ranges) < math_ops.to_int64(lengths))
  return masks 

def _length_penalty(sequence_lengths, penalty_factor):
  """Calculates the length penalty. See https://arxiv.org/abs/1609.08144.

  Args:
    sequence_lengths: The sequence length of all hypotheses, a tensor
      of shape [beam_size, vocab_size].
    penalty_factor: A scalar that weights the length penalty.

  Returns:
    The length penalty factor, a tensor fo shape [beam_size].
  """
  penalty_factor = ops.convert_to_tensor(penalty_factor, name="penalty_factor")
  penalty_factor.set_shape(())  # penalty should be a scalar.
  static_penalty = tensor_util.constant_value(penalty_factor)
  if static_penalty is not None and static_penalty == 0:
    return 1.0
  return math_ops.div((5. + math_ops.to_float(sequence_lengths))
                      **penalty_factor, (5. + 1.)**penalty_factor) 

def loss(self, data, labels):
    """The loss to minimize while training."""

    if self.is_regression:
      diff = self.training_inference_graph(data) - math_ops.to_float(labels)
      mean_squared_error = math_ops.reduce_mean(diff * diff)
      root_mean_squared_error = math_ops.sqrt(mean_squared_error, name="loss")
      loss = root_mean_squared_error
    else:
      loss = math_ops.reduce_mean(
          nn_ops.sparse_softmax_cross_entropy_with_logits(
              labels=array_ops.squeeze(math_ops.to_int32(labels)),
              logits=self.training_inference_graph(data)),
          name="loss")
    if self.regularizer:
      loss += layers.apply_regularization(self.regularizer,
                                          variables.trainable_variables())
    return loss 

def _lengths_to_masks(lengths, max_length):
  """Creates a binary matrix that can be used to mask away padding.

  Args:
    lengths: A vector of integers representing lengths.
    max_length: An integer indicating the maximum length. All values in
      lengths should be less than max_length.
  Returns:
    masks: Masks that can be used to get rid of padding.
  """
  tiled_ranges = array_ops.tile(
      array_ops.expand_dims(math_ops.range(max_length), 0),
      [array_ops.shape(lengths)[0], 1])
  lengths = array_ops.expand_dims(lengths, 1)
  masks = math_ops.to_float(
      math_ops.to_int64(tiled_ranges) < math_ops.to_int64(lengths))
  return masks 

def loss(self, data, labels):
    """The loss to minimize while training."""

    if self.is_regression:
      diff = self.training_inference_graph(data) - math_ops.to_float(labels)
      mean_squared_error = math_ops.reduce_mean(diff * diff)
      root_mean_squared_error = math_ops.sqrt(mean_squared_error, name="loss")
      loss = root_mean_squared_error
    else:
      loss = math_ops.reduce_mean(
          nn_ops.sparse_softmax_cross_entropy_with_logits(
              labels=array_ops.squeeze(math_ops.to_int32(labels)),
              logits=self.training_inference_graph(data)),
          name="loss")
    if self.regularizer:
      loss += layers.apply_regularization(self.regularizer,
                                          variables.trainable_variables())
    return loss 

def _define_full_covariance_probs(self, shard_id, shard):
    """Defines the full covariance probabilties per example in a class.

    Updates a matrix with dimension num_examples X num_classes.

    Args:
      shard_id: id of the current shard.
      shard: current data shard, 1 X num_examples X dimensions.
    """
    diff = shard - self._means
    cholesky = linalg_ops.cholesky(self._covs + self._min_var)
    log_det_covs = 2.0 * math_ops.reduce_sum(
        math_ops.log(array_ops.matrix_diag_part(cholesky)), 1)
    x_mu_cov = math_ops.square(
        linalg_ops.matrix_triangular_solve(
            cholesky, array_ops.transpose(
                diff, perm=[0, 2, 1]), lower=True))
    diag_m = array_ops.transpose(math_ops.reduce_sum(x_mu_cov, 1))
    self._probs[shard_id] = -0.5 * (diag_m + math_ops.to_float(self._dimensions)
                                    * math_ops.log(2 * np.pi) + log_det_covs) 

def _covariance(x, diag):
  """Defines the covariance operation of a matrix.

  Args:
    x: a matrix Tensor. Dimension 0 should contain the number of examples.
    diag: if True, it computes the diagonal covariance.

  Returns:
    A Tensor representing the covariance of x. In the case of
  diagonal matrix just the diagonal is returned.
  """
  num_points = math_ops.to_float(array_ops.shape(x)[0])
  x -= math_ops.reduce_mean(x, 0, keep_dims=True)
  if diag:
    cov = math_ops.reduce_sum(
        math_ops.square(x), 0, keep_dims=True) / (num_points - 1)
  else:
    cov = math_ops.matmul(x, x, transpose_a=True) / (num_points - 1)
  return cov 

def _define_full_covariance_probs(self, shard_id, shard):
    """Defines the full covariance probabilties per example in a class.

    Updates a matrix with dimension num_examples X num_classes.

    Args:
      shard_id: id of the current shard.
      shard: current data shard, 1 X num_examples X dimensions.
    """
    diff = shard - self._means
    cholesky = linalg_ops.cholesky(self._covs + self._min_var)
    log_det_covs = 2.0 * math_ops.reduce_sum(
        math_ops.log(array_ops.matrix_diag_part(cholesky)), 1)
    x_mu_cov = math_ops.square(
        linalg_ops.matrix_triangular_solve(
            cholesky, array_ops.transpose(
                diff, perm=[0, 2, 1]), lower=True))
    diag_m = array_ops.transpose(math_ops.reduce_sum(x_mu_cov, 1))
    self._probs[shard_id] = -0.5 * (diag_m + math_ops.to_float(self._dimensions)
                                    * math_ops.log(2 * np.pi) + log_det_covs) 

def create_test_input(batch_size, height, width, channels):
  """Create test input tensor.

  Args:
    batch_size: The number of images per batch or `None` if unknown.
    height: The height of each image or `None` if unknown.
    width: The width of each image or `None` if unknown.
    channels: The number of channels per image or `None` if unknown.

  Returns:
    Either a placeholder `Tensor` of dimension
      [batch_size, height, width, channels] if any of the inputs are `None` or a
    constant `Tensor` with the mesh grid values along the spatial dimensions.
  """
  if None in [batch_size, height, width, channels]:
    return array_ops.placeholder(dtypes.float32,
                                 (batch_size, height, width, channels))
  else:
    return math_ops.to_float(
        np.tile(
            np.reshape(
                np.reshape(np.arange(height), [height, 1]) + np.reshape(
                    np.arange(width), [1, width]), [1, height, width, 1]),
            [batch_size, 1, 1, channels])) 

def _covariance(x, diag):
  """Defines the covariance operation of a matrix.

  Args:
    x: a matrix Tensor. Dimension 0 should contain the number of examples.
    diag: if True, it computes the diagonal covariance.

  Returns:
    A Tensor representing the covariance of x. In the case of
  diagonal matrix just the diagonal is returned.
  """
  num_points = math_ops.to_float(array_ops.shape(x)[0])
  x -= math_ops.reduce_mean(x, 0, keep_dims=True)
  if diag:
    cov = math_ops.reduce_sum(
        math_ops.square(x), 0, keep_dims=True) / (num_points - 1)
  else:
    cov = math_ops.matmul(x, x, transpose_a=True) / (num_points - 1)
  return cov 

def _streaming_auc(predictions, labels, weights=None, class_id=None,
                   curve="ROC"):
  # pylint: disable=missing-docstring
  predictions = math_ops.to_float(predictions)
  if labels.dtype.base_dtype != dtypes.bool:
    logging.warning("Casting %s labels to bool.", labels.dtype)
    labels = math_ops.cast(labels, dtypes.bool)
  weights = _float_weights_or_none(weights)
  if weights is not None:
    weights = weights_broadcast_ops.broadcast_weights(weights, predictions)
  if class_id is not None:
    if weights is not None:
      weights = weights[:, class_id]
    predictions = predictions[:, class_id]
    labels = labels[:, class_id]
  return metrics_lib.streaming_auc(
      predictions, labels, weights=weights, curve=curve) 

def compute_weighted_loss(losses, weights=1.0, scope=None):
  """Computes the weighted loss.

  Args:
    losses: A tensor of size [batch_size, d1, ... dN].
    weights: A tensor of size [1] or [batch_size, d1, ... dK] where K < N.
    scope: the scope for the operations performed in computing the loss.

  Returns:
    A scalar `Tensor` that returns the weighted loss.

  Raises:
    ValueError: If `weights` is `None` or the shape is not compatible with
      `losses`, or if the number of dimensions (rank) of either `losses` or
      `weights` is missing.
  """
  with ops.name_scope(scope, "weighted_loss", [losses, weights]):
    losses = ops.convert_to_tensor(losses)
    input_dtype = losses.dtype
    losses = math_ops.to_float(losses)
    weights = math_ops.to_float(ops.convert_to_tensor(weights))

    if losses.get_shape().ndims is None:
      raise ValueError("losses.get_shape().ndims cannot be None")
    weights_shape = weights.get_shape()
    if weights_shape.ndims is None:
      raise ValueError("weights.get_shape().ndims cannot be None")

    if weights_shape.ndims > 1 and weights_shape.dims[-1].is_compatible_with(1):
      weights = array_ops.squeeze(weights, [-1])

    total_loss = _scale_losses(losses, weights)
    num_present = _num_present(losses, weights)
    mean_loss = _safe_mean(total_loss, num_present)
    # convert the result back to the input type
    mean_loss = math_ops.cast(mean_loss, input_dtype)
    add_loss(mean_loss)
    return mean_loss 

def compute_weighted_loss(
    losses, weights=1.0, scope=None, loss_collection=ops.GraphKeys.LOSSES):
  """Computes the weighted loss.

  Args:
    losses: `Tensor` of shape `[batch_size, d1, ... dN]`.
    weights: Optional `Tensor` whose rank is either 0, or the same rank as
      `losses`, and must be broadcastable to `losses` (i.e., all dimensions must
      be either `1`, or the same as the corresponding `losses` dimension).
    scope: the scope for the operations performed in computing the loss.
    loss_collection: the loss will be added to these collections.

  Returns:
    A scalar `Tensor` that returns the weighted loss.

  Raises:
    ValueError: If `weights` is `None` or the shape is not compatible with
      `losses`, or if the number of dimensions (rank) of either `losses` or
      `weights` is missing.
  """
  with ops.name_scope(scope, "weighted_loss", (losses, weights)):
    with ops.control_dependencies((
        weights_broadcast_ops.assert_broadcastable(weights, losses),)):
      losses = ops.convert_to_tensor(losses)
      input_dtype = losses.dtype
      losses = math_ops.to_float(losses)
      weights = math_ops.to_float(weights)
      total_loss = _scale_losses(losses, weights)
      num_present = _num_present(losses, weights)
      mean_loss = _safe_mean(total_loss, num_present)
      # Convert the result back to the input type.
      mean_loss = math_ops.cast(mean_loss, input_dtype)
      util.add_loss(mean_loss, loss_collection)
      return mean_loss 

def testConv2DSameOdd(self):
    n, n2 = 5, 3

    # Input image.
    x = create_test_input(1, n, n, 1)

    # Convolution kernel.
    w = create_test_input(1, 3, 3, 1)
    w = array_ops.reshape(w, [3, 3, 1, 1])

    variable_scope.get_variable('Conv/weights', initializer=w)
    variable_scope.get_variable('Conv/biases', initializer=array_ops.zeros([1]))
    variable_scope.get_variable_scope().reuse_variables()

    y1 = layers.conv2d(x, 1, [3, 3], stride=1, scope='Conv')
    y1_expected = math_ops.to_float([[14, 28, 43, 58, 34],
                                     [28, 48, 66, 84, 46],
                                     [43, 66, 84, 102, 55],
                                     [58, 84, 102, 120, 64],
                                     [34, 46, 55, 64, 30]])
    y1_expected = array_ops.reshape(y1_expected, [1, n, n, 1])

    y2 = resnet_utils.subsample(y1, 2)
    y2_expected = math_ops.to_float([[14, 43, 34],
                                     [43, 84, 55],
                                     [34, 55, 30]])
    y2_expected = array_ops.reshape(y2_expected, [1, n2, n2, 1])

    y3 = resnet_utils.conv2d_same(x, 1, 3, stride=2, scope='Conv')
    y3_expected = y2_expected

    y4 = layers.conv2d(x, 1, [3, 3], stride=2, scope='Conv')
    y4_expected = y2_expected

    with self.test_session() as sess:
      sess.run(variables.global_variables_initializer())
      self.assertAllClose(y1.eval(), y1_expected.eval())
      self.assertAllClose(y2.eval(), y2_expected.eval())
      self.assertAllClose(y3.eval(), y3_expected.eval())
      self.assertAllClose(y4.eval(), y4_expected.eval()) 

def testConv2DSameEven(self):
    n, n2 = 4, 2

    # Input image.
    x = create_test_input(1, n, n, 1)

    # Convolution kernel.
    w = create_test_input(1, 3, 3, 1)
    w = array_ops.reshape(w, [3, 3, 1, 1])

    variable_scope.get_variable('Conv/weights', initializer=w)
    variable_scope.get_variable('Conv/biases', initializer=array_ops.zeros([1]))
    variable_scope.get_variable_scope().reuse_variables()

    y1 = layers.conv2d(x, 1, [3, 3], stride=1, scope='Conv')
    y1_expected = math_ops.to_float([[14, 28, 43, 26], [28, 48, 66, 37],
                                     [43, 66, 84, 46], [26, 37, 46, 22]])
    y1_expected = array_ops.reshape(y1_expected, [1, n, n, 1])

    y2 = resnet_utils.subsample(y1, 2)
    y2_expected = math_ops.to_float([[14, 43], [43, 84]])
    y2_expected = array_ops.reshape(y2_expected, [1, n2, n2, 1])

    y3 = resnet_utils.conv2d_same(x, 1, 3, stride=2, scope='Conv')
    y3_expected = y2_expected

    y4 = layers.conv2d(x, 1, [3, 3], stride=2, scope='Conv')
    y4_expected = math_ops.to_float([[48, 37], [37, 22]])
    y4_expected = array_ops.reshape(y4_expected, [1, n2, n2, 1])

    with self.test_session() as sess:
      sess.run(variables.global_variables_initializer())
      self.assertAllClose(y1.eval(), y1_expected.eval())
      self.assertAllClose(y2.eval(), y2_expected.eval())
      self.assertAllClose(y3.eval(), y3_expected.eval())
      self.assertAllClose(y4.eval(), y4_expected.eval()) 

def log_loss(labels, predictions, weights=1.0, epsilon=1e-7, scope=None,
             loss_collection=ops.GraphKeys.LOSSES):
  """Adds a Log Loss term to the training procedure.

  `weights` acts as a coefficient for the loss. If a scalar is provided, then
  the loss is simply scaled by the given value. If `weights` is a tensor of size
  [batch_size], then the total loss for each sample of the batch is rescaled
  by the corresponding element in the `weights` vector. If the shape of
  `weights` matches the shape of `predictions`, then the loss of each
  measurable element of `predictions` is scaled by the corresponding value of
  `weights`.

  Args:
    labels: The ground truth output tensor, same dimensions as 'predictions'.
    predictions: The predicted outputs.
    weights: Optional `Tensor` whose rank is either 0, or the same rank as
      `labels`, and must be broadcastable to `labels` (i.e., all dimensions must
      be either `1`, or the same as the corresponding `losses` dimension).
    epsilon: A small increment to add to avoid taking a log of zero.
    scope: The scope for the operations performed in computing the loss.
    loss_collection: collection to which the loss will be added.

  Returns:
    A scalar `Tensor` representing the loss value.

  Raises:
    ValueError: If the shape of `predictions` doesn't match that of `labels` or
      if the shape of `weights` is invalid.
  """
  with ops.name_scope(scope, "log_loss",
                      (predictions, labels, weights)) as scope:
    predictions = math_ops.to_float(predictions)
    labels = math_ops.to_float(labels)
    predictions.get_shape().assert_is_compatible_with(labels.get_shape())
    losses = -math_ops.multiply(
        labels,
        math_ops.log(predictions + epsilon)) - math_ops.multiply(
            (1 - labels), math_ops.log(1 - predictions + epsilon))
    return compute_weighted_loss(losses, weights, scope, loss_collection) 

def hinge_loss(labels, logits, weights=1.0, scope=None,
               loss_collection=ops.GraphKeys.LOSSES):
  """Adds a hinge loss to the training procedure.

  Args:
    labels: The ground truth output tensor. Its shape should match the shape of
      logits. The values of the tensor are expected to be 0.0 or 1.0.
    logits: The logits, a float tensor.
    weights: Optional `Tensor` whose rank is either 0, or the same rank as
      `labels`, and must be broadcastable to `labels` (i.e., all dimensions must
      be either `1`, or the same as the corresponding `losses` dimension).
    scope: The scope for the operations performed in computing the loss.
    loss_collection: collection to which the loss will be added.

  Returns:
    A scalar `Tensor` of the loss value.

  Raises:
    ValueError: If the shapes of `logits` and `labels` don't match.
  """
  with ops.name_scope(scope, "hinge_loss", (logits, labels)) as scope:
    logits = math_ops.to_float(logits)
    labels = math_ops.to_float(labels)
    logits.get_shape().assert_is_compatible_with(labels.get_shape())
    # We first need to convert binary labels to -1/1 labels (as floats).
    all_ones = array_ops.ones_like(labels)
    labels = math_ops.subtract(2 * labels, all_ones)
    losses = nn_ops.relu(
        math_ops.subtract(all_ones, math_ops.multiply(labels, logits)))
    return compute_weighted_loss(losses, weights, scope, loss_collection) 

def cosine_distance(
    labels, predictions, dim=None, weights=1.0, scope=None,
    loss_collection=ops.GraphKeys.LOSSES):
  """Adds a cosine-distance loss to the training procedure.

  Note that the function assumes that `predictions` and `labels` are already
  unit-normalized.

  Args:
    labels: `Tensor` whose shape matches 'predictions'
    predictions: An arbitrary matrix.
    dim: The dimension along which the cosine distance is computed.
    weights: Optional `Tensor` whose rank is either 0, or the same rank as
      `labels`, and must be broadcastable to `labels` (i.e., all dimensions must
      be either `1`, or the same as the corresponding `losses` dimension).
    scope: The scope for the operations performed in computing the loss.
    loss_collection: collection to which this loss will be added.

  Returns:
    A scalar `Tensor` representing the loss value.

  Raises:
    ValueError: If `predictions` shape doesn't match `labels` shape, or
      `weights` is `None`.
  """
  if dim is None:
    raise ValueError("`dim` cannot be None.")
  with ops.name_scope(scope, "cosine_distance_loss",
                      (predictions, labels, weights)) as scope:
    predictions = math_ops.to_float(predictions)
    labels = math_ops.to_float(labels)
    predictions.get_shape().assert_is_compatible_with(labels.get_shape())

    radial_diffs = math_ops.multiply(predictions, labels)
    losses = 1 - math_ops.reduce_sum(radial_diffs, axis=(dim,), keep_dims=True)
    return compute_weighted_loss(losses, weights, scope, loss_collection) 

def mean_squared_error(labels, predictions, weights=1.0, scope=None,
                       loss_collection=ops.GraphKeys.LOSSES):
  """Adds a Sum-of-Squares loss to the training procedure.

  `weights` acts as a coefficient for the loss. If a scalar is provided, then
  the loss is simply scaled by the given value. If `weights` is a tensor of size
  [batch_size], then the total loss for each sample of the batch is rescaled
  by the corresponding element in the `weights` vector. If the shape of
  `weights` matches the shape of `predictions`, then the loss of each
  measurable element of `predictions` is scaled by the corresponding value of
  `weights`.

  Args:
    labels: The ground truth output tensor, same dimensions as 'predictions'.
    predictions: The predicted outputs.
    weights: Optional `Tensor` whose rank is either 0, or the same rank as
      `labels`, and must be broadcastable to `labels` (i.e., all dimensions must
      be either `1`, or the same as the corresponding `losses` dimension).
    scope: The scope for the operations performed in computing the loss.
    loss_collection: collection to which the loss will be added.

  Returns:
    A scalar `Tensor` representing the loss value.

  Raises:
    ValueError: If the shape of `predictions` doesn't match that of `labels` or
      if the shape of `weights` is invalid.
  """
  with ops.name_scope(scope, "mean_squared_error",
                      (predictions, labels, weights)) as scope:
    predictions = math_ops.to_float(predictions)
    labels = math_ops.to_float(labels)
    predictions.get_shape().assert_is_compatible_with(labels.get_shape())
    losses = math_ops.square(math_ops.subtract(predictions, labels))
    return compute_weighted_loss(losses, weights, scope, loss_collection) 

def _log_loss_with_two_classes(logits, target):
  # sigmoid_cross_entropy_with_logits requires [batch_size, 1] target.
  if len(target.get_shape()) == 1:
    target = array_ops.expand_dims(target, dim=[1])
  loss_vec = nn.sigmoid_cross_entropy_with_logits(
      labels=math_ops.to_float(target), logits=logits)
  return loss_vec 

def _streaming_accuracy_at_threshold(predictions, labels, weights, threshold):
  threshold_predictions = math_ops.to_float(
      math_ops.greater_equal(predictions, threshold))
  return metrics_lib.streaming_accuracy(
      predictions=threshold_predictions, labels=labels, weights=weights) 

def absolute_difference(predictions, labels=None, weights=1.0, scope=None):
  """Adds an Absolute Difference loss to the training procedure.

  `weights` acts as a coefficient for the loss. If a scalar is provided, then
  the loss is simply scaled by the given value. If `weights` is a tensor of size
  [batch_size], then the total loss for each sample of the batch is rescaled
  by the corresponding element in the `weights` vector. If the shape of
  `weights` matches the shape of `predictions`, then the loss of each
  measurable element of `predictions` is scaled by the corresponding value of
  `weights`.

  Args:
    predictions: The predicted outputs.
    labels: The ground truth output tensor, same dimensions as 'predictions'.
    weights: Coefficients for the loss a scalar, a tensor of shape
      [batch_size] or a tensor whose shape matches `predictions`.
    scope: The scope for the operations performed in computing the loss.

  Returns:
    A scalar `Tensor` representing the loss value.

  Raises:
    ValueError: If the shape of `predictions` doesn't match that of `labels` or
      if the shape of `weights` is invalid.
  """
  with ops.name_scope(scope, "absolute_difference",
                      [predictions, labels, weights]) as scope:
    predictions.get_shape().assert_is_compatible_with(labels.get_shape())
    predictions = math_ops.to_float(predictions)
    labels = math_ops.to_float(labels)
    losses = math_ops.abs(math_ops.subtract(predictions, labels))
    return compute_weighted_loss(losses, weights, scope=scope) 

def _to_dnn_input_layer(self,
                          input_tensor,
                          weight_collections=None,
                          trainable=True,
                          output_rank=2):
    if input_tensor.dtype != dtypes.float32:
      input_tensor = math_ops.to_float(input_tensor)
    return _reshape_real_valued_tensor(input_tensor, output_rank, self.name) 

def _float_weights_or_none(weights):
  if weights is None:
    return None
  return math_ops.to_float(weights) 

def _adaptive_max_norm(norm, std_factor, decay, global_step, epsilon, name):
  """Find max_norm given norm and previous average."""
  with vs.variable_scope(name, "AdaptiveMaxNorm", [norm]):
    log_norm = math_ops.log(norm + epsilon)

    def moving_average(name, value, decay):
      moving_average_variable = vs.get_variable(
          name,
          shape=value.get_shape(),
          dtype=value.dtype,
          initializer=init_ops.zeros_initializer(),
          trainable=False)
      return moving_averages.assign_moving_average(
          moving_average_variable, value, decay, zero_debias=False)

    # quicker adaptation at the beginning
    if global_step is not None:
      n = math_ops.to_float(global_step)
      decay = math_ops.minimum(decay, n / (n + 1.))

    # update averages
    mean = moving_average("mean", log_norm, decay)
    sq_mean = moving_average("sq_mean", math_ops.square(log_norm), decay)

    variance = sq_mean - math_ops.square(mean)
    std = math_ops.sqrt(math_ops.maximum(epsilon, variance))
    max_norms = math_ops.exp(mean + std_factor * std)
    return max_norms, mean 

def _to_dnn_input_layer(self,
                          input_tensor,
                          weight_collections=None,
                          trainable=True,
                          output_rank=2):
    input_tensor = self._to_dense_tensor(input_tensor)
    if input_tensor.dtype != dtypes.float32:
      input_tensor = math_ops.to_float(input_tensor)
    return _reshape_real_valued_tensor(input_tensor, output_rank, self.name) 

def insert_transformed_feature(self, columns_to_tensors):
    """Apply transformation and inserts it into columns_to_tensors.

    Args:
      columns_to_tensors: A mapping from feature columns to tensors. 'string'
        key means a base feature (not-transformed). It can have _FeatureColumn
        as a key too. That means that _FeatureColumn is already transformed.
    """
    # Transform the input tensor according to the normalizer function.
    input_tensor = self._normalized_input_tensor(columns_to_tensors[self.name])
    columns_to_tensors[self] = math_ops.to_float(input_tensor)

  # pylint: disable=unused-argument 

def insert_transformed_feature(self, columns_to_tensors):
    """Apply transformation and inserts it into columns_to_tensors.

    Args:
      columns_to_tensors: A mapping from feature columns to tensors. 'string'
        key means a base feature (not-transformed). It can have _FeatureColumn
        as a key too. That means that _FeatureColumn is already transformed.
    """
    # Transform the input tensor according to the normalizer function.
    input_tensor = self._normalized_input_tensor(columns_to_tensors[self.name])
    columns_to_tensors[self] = math_ops.to_float(input_tensor)

  # pylint: disable=unused-argument 

def _do_transform(self, id_tensor, weight_tensor):
    if not isinstance(weight_tensor, sparse_tensor_py.SparseTensor):
      # The weight tensor can be a regular Tensor. In such case, sparsify it.
      weight_tensor = contrib_sparse_ops.dense_to_sparse_tensor(weight_tensor)
    if not self.dtype.is_floating:
      weight_tensor = math_ops.to_float(weight_tensor)
    return tuple([id_tensor, weight_tensor])