Function l2_normalize_v2

tensorflow/python/ops/nn_impl.py:623–653

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

(x, axis=None, epsilon=1e-12, name=None, do_fusion=True)

@tf_export("math.l2_normalize", "linalg.l2_normalize", "nn.l2_normalize", v1=[])
def l2_normalize_v2(x, axis=None, epsilon=1e-12, name=None, do_fusion=True):
  """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)`.
    do_fusion: Whether fuse op when doing l2 norm on last axis, enabled by default.
    name: A name for this operation (optional).

  Returns:
    A `Tensor` with the same shape as `x`.
  """
  if do_fusion and x.dtype == dtypes.float32 and (
        axis is None or axis== x.shape.rank - 1):
    return fused_l2_normalize(x, epsilon=epsilon, name=name)
  else:
    with ops.name_scope(name, "l2_normalize", [x]) as name:
      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)

def fused_l2_normalize(x, epsilon=1e-12, name=None):
  """Normalizes along last dimension using an L2 norm.