mindspore.mint.nn.functional.linear

mindspore.mint.nn.functional.linear(input, weight, bias=None)

Applies the dense connected operation to the input. The dense function is defined as:

$$output=input\ast weigh{t}^T+bias$$

Warning

• This is an experimental API that is subject to change or deletion.

• On the Ascend platform, if bias is not 1D, the input cannot be greater than 6D in PYNATIVE or KBK mode.

Parameters

• input (Tensor) – Input Tensor of shape $(\ast ,in\mathrm{\_}channels)$, where $\ast$ means any number of additional dimensions.

• weight (Tensor) – The weight applied to the input. The shape is $(out\mathrm{\_}channels,in\mathrm{\_}channels)$ or $(in\mathrm{\_}channels)$.

• bias (Tensor, optional) – Additive biases to the output. The shape is $(out\mathrm{\_}channels)$ or $()$. Defaults: None, the bias is 0.

Returns

Output whose shape is determined by the shape of the input and the weight.

Raises

• TypeError – If input is not Tensor.

• TypeError – If weight is not Tensor.

• TypeError – If bias is not Tensor.

• RuntimeError – On the Ascend platform, if bias is not 1D and input is greater than 6D in PYNATIVE or KBK mode.

Supported Platforms:

Ascend

Examples

>>> import numpy as np
>>> import mindspore
>>> from mindspore import Tensor, mint
>>> input = Tensor([[-1., 1., 2.], [-3., -3., 1.]], mindspore.float32)
>>> weight = Tensor([[-2., -2., -2.], [0., -1., 0.]], mindspore.float32)
>>> bias = Tensor([0., 1.], mindspore.float32)
>>> output = mint.nn.functional.linear(input, weight, bias)
>>> print(output)
[[-4.  0.]
 [10.  4.]]