【Tensorflow Learning 】nn.l2_normalize

Learning Some thing about nn.l2_normalize , the formali in function declare as :
output = x / sqrt(max(sum(x**2), epsilon))
which could be declare as :
output = x / |x|

@tf_export("nn.l2_normalize")
@deprecated_args(None, "dim is deprecated, use axis instead", "dim")
def l2_normalize(x, axis=None, epsilon=1e-12, name=None, dim=None):
  """Normalizes along dimension `axis` using an L2 norm.

  For a 1-D tensor with `axis = 0`, computes

      output = x / sqrt(max(sum(x**2), epsilon))

  For `x` with more dimensions, independently normalizes each 1-D slice along
  dimension `axis`.

  Args:
    x: A `Tensor`.
    axis: Dimension along which to normalize.  A scalar or a vector of
      integers.
    epsilon: A lower bound value for the norm. Will use `sqrt(epsilon)` as the
      divisor if `norm < sqrt(epsilon)`.
    name: A name for this operation (optional).
    dim: Deprecated alias for axis.

  Returns:
    A `Tensor` with the same shape as `x`.
  """
  with ops.name_scope(name, "l2_normalize", [x]) as name:
    axis = deprecated_argument_lookup("axis", axis, "dim", dim)
    x = ops.convert_to_tensor(x, name="x")
    square_sum = math_ops.reduce_sum(math_ops.square(x), axis, keepdims=True)
    x_inv_norm = math_ops.rsqrt(math_ops.maximum(square_sum, epsilon))
    return math_ops.multiply(x, x_inv_norm, name=name)