[ 参数不一致 ]torch.nn.functional.batch_norm¶

torch.nn.functional.batch_norm¶

torch.nn.functional.batch_norm(input, running_mean, running_var, weight=None, bias=None, training=False, momentum=0.1, eps=1e-05)

paddle.nn.functional.batch_norm¶

paddle.nn.functional.batch_norm(x, running_mean, running_var, weight, bias, training=False, momentum=0.9, epsilon=1e-05, data_format='NCHW', name=None)

其中 Pytorch 与 Paddle 参数不一致，具体如下：

参数映射¶

| PyTorch | PaddlePaddle | 备注 | | ————- | ———— | —————————————————— | | input | x | 表示输入的 Tensor ，仅参数名不一致。 | | running_mean | running_mean | 均值的 Tensor | | running_var | running_var | 方差的 Tensor | | weight | weight | 权重的 Tensor | | bias | bias | 偏置的 Tensor | | eps | epsilon | 为了数值稳定加在分母上的值 | | momentum | momentum | 此值用于计算 moving_mean 和 moving_var, 值的大小 Pytorch = 1 - Paddle，需要进行转写 | | training | training | 是否可训练。 | | - | data_format | 指定输入数据格式。 Pytorch 无此参数。保持默认即可。 |

转写示例¶

# Pytorch 写法
torch.nn.functional.batch_norm(input=input, running_mean=running_mean, running_var=running_var, momentum=0.1)

# Paddle 写法
paddle.nn.functional.batch_norm(x=input, running_mean=running_mean, running_var=running_var, momentum=0.9)