Identity

Contents

Identity#

class braintools.init.Identity(scale=1.0, unit=Unit('1'))#

Identity matrix initialization.

Initializes weights to an identity matrix, optionally scaled by a scale factor. For non-square matrices, creates a matrix that is as close to identity as possible.

Parameters:

scale (float) – Multiplicative factor to apply to the identity matrix (default: 1.0).

Notes

The output depends on the rank of size:

  • 1-D (n,): an identity matrix is undefined, so a vector of scale * ones(n) is returned (useful as a bias-like / pass-through init).

  • 2-D (rows, cols): scale * eye(rows, cols) (identity, truncated or zero-padded for rectangular shapes).

  • N-D: scale times an identity placed in the trailing two dimensions, broadcast over the leading dimensions.

Examples

>>> import numpy as np
>>> from braintools.init import Identity
>>>
>>> init = Identity(scale=1.0)
>>> weights = init((100, 100))
>>>
>>> # For rectangular matrices
>>> weights = init((100, 50))  # Will create identity-like matrix
Contents