PyTorch：损失函数loss function

偏执的太偏执、 2022-12-16 15:24 120阅读 0赞

[\-柚子皮-][-_-]

# [Loss Functions][] #

<table> 
 <tbody> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.L1Loss.html#torch.nn.L1Loss" rel="nofollow"><code>nn.L1Loss</code></a> </td> 
 <td style="width:601px;"> Creates a criterion that measures the mean absolute error (MAE) between each element in the input&nbsp;xx&nbsp;and target&nbsp;yy&nbsp;. </td> 
 </tr> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.MSELoss.html#torch.nn.MSELoss" rel="nofollow"><code>nn.MSELoss</code></a> </td> 
 <td style="width:601px;"> Creates a criterion that measures the mean squared error (squared L2 norm) between each element in the input&nbsp;xx&nbsp;and target&nbsp;yy&nbsp;. </td> 
 </tr> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.CrossEntropyLoss.html#torch.nn.CrossEntropyLoss" rel="nofollow"><code>nn.CrossEntropyLoss</code></a> </td> 
 <td style="width:601px;"> This criterion combines&nbsp;<a href="https://pytorch.org/docs/stable/generated/torch.nn.LogSoftmax.html#torch.nn.LogSoftmax" rel="nofollow"><code>nn.LogSoftmax()</code></a>&nbsp;and&nbsp;<a href="https://pytorch.org/docs/stable/generated/torch.nn.NLLLoss.html#torch.nn.NLLLoss" rel="nofollow"><code>nn.NLLLoss()</code></a>&nbsp;in one single class. </td> 
 </tr> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.CTCLoss.html#torch.nn.CTCLoss" rel="nofollow"><code>nn.CTCLoss</code></a> </td> 
 <td style="width:601px;"> The Connectionist Temporal Classification loss. </td> 
 </tr> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.NLLLoss.html#torch.nn.NLLLoss" rel="nofollow"><code>nn.NLLLoss</code></a> </td> 
 <td style="width:601px;"> The negative log likelihood loss. </td> 
 </tr> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.PoissonNLLLoss.html#torch.nn.PoissonNLLLoss" rel="nofollow"><code>nn.PoissonNLLLoss</code></a> </td> 
 <td style="width:601px;"> Negative log likelihood loss with Poisson distribution of target. </td> 
 </tr> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.KLDivLoss.html#torch.nn.KLDivLoss" rel="nofollow"><code>nn.KLDivLoss</code></a> </td> 
 <td style="width:601px;"> The Kullback-Leibler divergence loss measure </td> 
 </tr> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.BCELoss.html#torch.nn.BCELoss" rel="nofollow"><code>nn.BCELoss</code></a> </td> 
 <td style="width:601px;"> Creates a criterion that measures the Binary Cross Entropy between the target and the output: </td> 
 </tr> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.BCEWithLogitsLoss.html#torch.nn.BCEWithLogitsLoss" rel="nofollow"><code>nn.BCEWithLogitsLoss</code></a> </td> 
 <td style="width:601px;"> This loss combines a&nbsp;Sigmoid&nbsp;layer and the&nbsp;BCELoss&nbsp;in one single class. </td> 
 </tr> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.MarginRankingLoss.html#torch.nn.MarginRankingLoss" rel="nofollow"><code>nn.MarginRankingLoss</code></a> </td> 
 <td style="width:601px;"> Creates a criterion that measures the loss given inputs&nbsp;x1x1&nbsp;,&nbsp;x2x2&nbsp;, two 1D mini-batch&nbsp;Tensors, and a label 1D mini-batch tensor&nbsp;yy&nbsp;(containing 1 or -1). </td> 
 </tr> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.HingeEmbeddingLoss.html#torch.nn.HingeEmbeddingLoss" rel="nofollow"><code>nn.HingeEmbeddingLoss</code></a> </td> 
 <td style="width:601px;"> Measures the loss given an input tensor&nbsp;xx&nbsp;and a labels tensor&nbsp;yy&nbsp;(containing 1 or -1). </td> 
 </tr> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.MultiLabelMarginLoss.html#torch.nn.MultiLabelMarginLoss" rel="nofollow"><code>nn.MultiLabelMarginLoss</code></a> </td> 
 <td style="width:601px;"> Creates a criterion that optimizes a multi-class multi-classification hinge loss (margin-based loss) between input&nbsp;xx&nbsp;(a 2D mini-batch&nbsp;Tensor) and output&nbsp;yy&nbsp;(which is a 2D&nbsp;Tensor&nbsp;of target class indices). </td> 
 </tr> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.SmoothL1Loss.html#torch.nn.SmoothL1Loss" rel="nofollow"><code>nn.SmoothL1Loss</code></a> </td> 
 <td style="width:601px;"> Creates a criterion that uses a squared term if the absolute element-wise error falls below beta and an L1 term otherwise. </td> 
 </tr> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.SoftMarginLoss.html#torch.nn.SoftMarginLoss" rel="nofollow"><code>nn.SoftMarginLoss</code></a> </td> 
 <td style="width:601px;"> Creates a criterion that optimizes a two-class classification logistic loss between input tensor&nbsp;xx&nbsp;and target tensor&nbsp;yy&nbsp;(containing 1 or -1). </td> 
 </tr> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.MultiLabelSoftMarginLoss.html#torch.nn.MultiLabelSoftMarginLoss" rel="nofollow"><code>nn.MultiLabelSoftMarginLoss</code></a> </td> 
 <td style="width:601px;"> Creates a criterion that optimizes a multi-label one-versus-all loss based on max-entropy, between input&nbsp;xx&nbsp;and target&nbsp;yy&nbsp;of size&nbsp;(N, C)(N,C)&nbsp;. </td> 
 </tr> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.CosineEmbeddingLoss.html#torch.nn.CosineEmbeddingLoss" rel="nofollow"><code>nn.CosineEmbeddingLoss</code></a> </td> 
 <td style="width:601px;"> Creates a criterion that measures the loss given input tensors&nbsp;x_1x1&nbsp;,&nbsp;x_2x2&nbsp;and a&nbsp;Tensor&nbsp;label&nbsp;yy&nbsp;with values 1 or -1. </td> 
 </tr> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.MultiMarginLoss.html#torch.nn.MultiMarginLoss" rel="nofollow"><code>nn.MultiMarginLoss</code></a> </td> 
 <td style="width:601px;"> Creates a criterion that optimizes a multi-class classification hinge loss (margin-based loss) between input&nbsp;xx&nbsp;(a 2D mini-batch&nbsp;Tensor) and output&nbsp;yy&nbsp;(which is a 1D tensor of target class indices,&nbsp;0 \leq y \leq \text{x.size}(1)-10≤y≤x.size(1)−1&nbsp;): </td> 
 </tr> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.TripletMarginLoss.html#torch.nn.TripletMarginLoss" rel="nofollow"><code>nn.TripletMarginLoss</code></a> </td> 
 <td style="width:601px;"> Creates a criterion that measures the triplet loss given an input tensors&nbsp;x1x1&nbsp;,&nbsp;x2x2&nbsp;,&nbsp;x3x3&nbsp;and a margin with a value greater than&nbsp;00&nbsp;. </td> 
 </tr> 
 <tr> 
 <td style="width:197px;"> <a href="https://pytorch.org/docs/stable/generated/torch.nn.TripletMarginWithDistanceLoss.html#torch.nn.TripletMarginWithDistanceLoss" rel="nofollow"><code>nn.TripletMarginWithDistanceLoss</code></a> </td> 
 <td style="width:601px;"> Creates a criterion that measures the triplet loss given input tensors&nbsp;aa&nbsp;,&nbsp;pp&nbsp;, and&nbsp;nn&nbsp;(representing anchor, positive, and negative examples, respectively), and a nonnegative, real-valued function (“distance function”) used to compute the relationship between the anchor and positive example (“positive distance”) and the anchor and negative example (“negative distance”). </td> 
 </tr> 
 </tbody> 
</table>

参数

**reduction** (*string**, **optional*) – Specifies the reduction to apply to the output: `'none'` | `'mean'` | `'sum'`. `'none'`: no reduction will be applied, `'mean'`: the weighted mean of the output is taken, `'sum'`: the output will be summed. Note: `size_average` and `reduce` are in the process of being deprecated, and in the meantime, specifying either of those two args will override `reduction`. Default: `'mean'`

\[[CROSSENTROPYLOSS][]\]

\[[pytorch loss function 总结][pytorch loss function]\]

示例：

import torch.nn.functional as F

labels = dataloader\["label"\]  
predictions = outputs.squeeze().contiguous()

loss = F.binary\_cross\_entropy(predictions, labels, reduction='mean')

from: [\-柚子皮-][-_-]

ref: \[[https://pytorch.org/docs/stable/nn.html\#loss-functions][https_pytorch.org_docs_stable_nn.html_loss-functions]\]

[-_-]: https://me.csdn.net/pipisorry
[Loss Functions]: https://pytorch.org/docs/stable/nn.html#id1
[CROSSENTROPYLOSS]: https://pytorch.org/docs/stable/generated/torch.nn.CrossEntropyLoss.html#torch.nn.CrossEntropyLoss
[pytorch loss function]: https://blog.csdn.net/zhangxb35/article/details/72464152
[https_pytorch.org_docs_stable_nn.html_loss-functions]: https://pytorch.org/docs/stable/nn.html#loss-functions