uniform_

→Tensor

uniform_(tensor: Tensor, a: float = 0.0, b: float = 1.0)

source

Initialise tensor in-place with samples from a uniform distribution.

Each entry is drawn independently from $\mathcal{U}(a, b)$ . The function mutates tensor in place and returns the same object so calls can be chained or used as a one-liner inside a model constructor.

Parameters

tensorTensor

Tensor to fill in place; any shape is accepted.

afloat= 0.0

Lower bound of the uniform interval. Default 0.0.

bfloat= 1.0

Upper bound of the uniform interval. Default 1.0.

Returns

Tensor

tensor (mutated) for chaining.

Notes

The resulting distribution has mean and variance

\mathbb{E}[W] = \frac{a + b}{2}, \qquad \mathrm{Var}(W) = \frac{(b - a)^2}{12}.

Higher-level schemes such as xavier_uniform_ and kaiming_uniform_ are thin wrappers over this function with bounds computed from the fan factors of tensor.

Examples

>>> import lucid
>>> from lucid.nn.init import uniform_
>>> w = lucid.empty(4, 4)
>>> uniform_(w, -0.1, 0.1)