tf.compat.v2.linalg.l2_normalize

View source on GitHub

Normalizes along dimension axis using an L2 norm.

Aliases:

• tf.compat.v2.math.l2_normalize
• tf.compat.v2.nn.l2_normalize

tf.compat.v2.linalg.l2_normalize(
    x,
    axis=None,
    epsilon=1e-12,
    name=None
)

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).

Returns:

A Tensor with the same shape as x.