\u200E

[torch 参数更多 ] torch.nn.functional.triplet_margin_loss¶

torch.nn.functional.triplet_margin_loss ¶

          torch.nn.functional.triplet_margin_loss(anchor,
                positive,
                negative,
                margin=1.0,
                p=2,
                eps=1e-06,
                swap=False,
                size_average=None,
                reduce=None,
                reduction='mean')

         

paddle.nn.functional.triplet_margin_loss ¶

          paddle.nn.functional.triplet_margin_loss(input,
                positive,
                negative,
                margin: float = 1.0,
                p: float = 2.0,
                epsilon: float = 1e-6,
                swap: bool = False,
                reduction: str = 'mean',
                name: str = None)

         

其中 Pytorch 相⽐ Paddle ⽀持更多其他参数，具体如下：

参数映射¶

| PyTorch | PaddlePaddle | 备注 | | ————- | ———— | —————————————————— | | anchor | input | 输入 Tensor，仅参数名不一致。 | | positive | positive | 输入正样本 | | negative | negative | 输入负样本 | | margin | margin | 手动指定间距 | | p | p | 指定范数 | | eps | epsilon | 防止除数为零的常数 | | swap | swap | 是否进行交换 | | size_average | - | 已弃用 | | reduce | - | 已弃用 | | reduction | reduction | 表示应用于输出结果的规约方式，可选值有：’none’, ‘mean’, ‘sum’ |

转写示例¶

size_average¶

           # 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'

          