[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'