tf.nn.l2_normalize

tf.nn.l2_normalize(
    x,
    axis=None,
    epsilon=1e-12,
    name=None,
    dim=None
)

Defined in tensorflow/python/ops/nn_impl.py.

See the guide: Neural Network > Normalization

Normalizes along dimension axis using an L2 norm. (deprecated arguments)

SOME ARGUMENTS ARE DEPRECATED. They will be removed in a future version. Instructions for updating: dim is deprecated, use axis instead

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.