mindspore.nn.MultiMarginLoss

class mindspore.nn.MultiMarginLoss(p=1, margin=1.0, reduction='mean', weight=None)[source]

Creates a criterion that optimizes a multi-class classification hinge loss (margin-based loss) between input \(x\) (a 2D mini-batch Tensor) and output \(y\) (which is a 1D tensor of target class indices, \(0 \leq y \leq \text{x.size}(1)-1\)):

For each mini-batch sample, the loss in terms of the 1D input \(x\) and scalar output \(y\) is:

\[\text{loss}(x, y) = \frac{\sum_i \max(0, w[y] * (\text{margin} - x[y] + x[i]))^p}{\text{x.size}(0)}\]

where \(x \in \left\{0, \; \cdots , \; \text{x.size}(0) - 1\right\}\) and \(i \neq y\).

Parameters

  • p (int, optional) – The norm degree for pairwise distance. Should be 1 or 2. Default: 1 .

  • margin (float, optional) – A parameter to change pairwise distance. Default: 1.0.

  • reduction (str, optional) –

    Apply specific reduction method to the output: 'none' , 'mean' , 'sum' . Default: 'mean' .

    • 'none': no reduction will be applied.

    • 'mean': compute and return the weighted mean of elements in the output.

    • 'sum': the output elements will be summed.

  • weight (Tensor, optional) – The rescaling weight to each class with shape \((C,)\). Data type only support float32, float16 or float64. Default: None , all classes are weighted equally.

Inputs:

  • x (Tensor) - Input x, with shape \((N, C)\). Data type only supports float32, float16 or float64. x is \(x\) in the above formula.

  • target (Tensor) - Ground truth labels, with shape \((N,)\). Data type only supports int64. The value of target should be non-negative, less than C. target is \(y\) in the above formula.

Outputs:

Tensor. When reduction is 'none', the shape is \((N,)\). Otherwise, it is a scalar. Has the same data type with x.

Raises

  • TypeError – If dtype of p or target is not int.

  • TypeError – If dtype of margin is not float.

  • TypeError – If dtype of reduction is not str.

  • TypeError – If dtype of x is not float16, float or float64.

  • TypeError – If dtype of weight and x is not the same.

  • ValueError – If p is not 1 or 2.

  • ValueError – If reduction is not one of { 'none' , 'sum' , 'mean' }.

  • ValueError – If shape[0] of x is not equal to shape[0] of target.

  • ValueError – If shape[1] of x is not equal to shape[0] of weight.

  • ValueError – IF rank of weight is not 1.

  • ValueError – If rank of x is not 2 or rank of 'target' is not 1.

Supported Platforms:

Ascend GPU CPU

Examples

>>> import mindspore as ms
>>> import mindspore.nn as nn
>>> import numpy as np
>>> x = ms.Tensor(np.ones(shape=[3, 3]), ms.float32)
>>> target = ms.Tensor(np.array([1, 2, 1]), ms.int64)
>>> loss = nn.MultiMarginLoss()
>>> output = loss(x, target)
>>> print(output)
0.6666667