\u200E

torch.nn.functional.binary_cross_entropy_with_logits¶

torch.nn.functional.binary_cross_entropy_with_logits ¶

          torch.nn.functional.binary_cross_entropy_with_logits(input, target, weight=None, size_average=None, reduce=None, reduction='mean', pos_weight=None)

         

paddle.nn.functional.binary_cross_entropy_with_logits ¶

          paddle.nn.functional.binary_cross_entropy_with_logits(logit, label, weight=None, reduction='mean', pos_weight=None, name=None)

         

两者功能一致，torch 参数多，具体如下：

参数差异¶

| PyTorch | PaddlePaddle | 备注 | | ————- | ———— | —————————————————— | | input | logit | 表示输入的 Tensor | | target | label | 标签 | | weight | weight | 类别权重 | | size_average | - | 已废弃，和 reduce 组合决定损失计算方式 | | reduce | - | 已废弃，和 size_average 组合决定损失计算方式 | | reduction | reduction | 输出结果的计算方式 | | pos_weight | pos_weight | 正类的权重 |

转写示例¶

          # Pytorch 的 size_average、reduce 参数转为 Paddle 的 reduction 参数
if size_average is None:
    size_average = True
if reduce is None:
    reduce = True

if size_average and reduce:
    reduction = 'mean'
elif reduce:
    reduction = 'sum'
else:
    reduction = 'none'

         