计算两个向量之间的距离，分别用nn.InstanceNorm1d 和手动实现

对于通道数只有1，长度为L 的向量，大小是 1*L ，计算这样的两个向量之间的距离，拓展到一个batch 共有n个样本向量，也即向量的大小是 n*1*L , 计算两个batch之间的距离（大小是[-1,1]）

import torch
import torch.nn as nn

from torch_geometric.data import DataLoader
def D1(p,z):
    norm = nn.InstanceNorm1d(1)
    p = norm(p)
    print(f'D1 norm P: {p}')
    # z = torch.instance_norm(z,dim=1)
    z = norm(z)
    print(f'D1 z: {z}')
    return -(p*z).sum(dim=1).mean()

def D2(p,z):
    # norm = nn.InstanceNorm1d(1)
    # manually mormalization
    N,C = p.size(0),p.size(1)
    p_t = p.reshape(N * C, -1)
    p_mean = p_t.mean(dim=1).reshape(N,C,-1)
    p_std = p_t.std(dim=1,unbiased=False).reshape(N,C,-1)
    p = (p-p_mean)/p_std
    print(f'p_mean: {p_mean},\n p_std: {p_std},\n D2 p_norm: {p}')

    z_t = z.reshape(N * C, -1)
    z_mean = z_t.mean(dim=1).reshape(N, C, -1)
    z_std = z_t.std(dim=1, unbiased=False).reshape(N, C, -1)
    z = (z - z_mean) / z_std
    print(f'z_mean: {z_mean}, \n z_std: {z_std},\n D2 z_norm: {z}')
    return -(p*z).sum(dim=1).mean()

if __name__ == '__main__':
    # similarity function D(p,z)
    torch.manual_seed(3)

    p = torch.randn(2,5)
    z = torch.randn(2, 5)
    print(f'p:{p}, \n z:{z}')

    N,C = p.size(0),p.size(1)
    p = p.reshape(N, 1, -1)

    z = z.reshape(N,1,-1)

    print(D1(p,z))
    print(D2(p,z))

两种方法计算结果是一样的：

p:tensor([[ 0.8033,  0.1748,  0.0890, -0.6137,  0.0462],
        [-1.3683,  0.3375,  1.0111, -1.4352,  0.9774]]), 
 z:tensor([[ 0.5220,  1.2379, -0.8646,  0.2990,  0.4192],
        [-0.0799,  0.9264,  0.8157,  0.4952, -0.1643]])
D1 norm P: tensor([[[ 1.5629,  0.1665, -0.0243, -1.5857, -0.1194]],

        [[-1.1640,  0.3960,  1.0120, -1.2252,  0.9812]]])
D1 z: tensor([[[ 0.2941,  1.3503, -1.7518, -0.0350,  0.1424]],

        [[-1.0658,  1.1756,  0.9290,  0.2151, -1.2539]]])
tensor(-0.1918)
p_mean: tensor([[[ 0.0999]],

        [[-0.0955]]]),
 p_std: tensor([[[0.4500]],

        [[1.0934]]]),
 D2 p_norm: tensor([[[ 1.5630,  0.1665, -0.0243, -1.5858, -0.1194]],

        [[-1.1640,  0.3960,  1.0120, -1.2252,  0.9812]]])
z_mean: tensor([[[0.3227]],

        [[0.3986]]]), 
 z_std: tensor([[[0.6778]],

        [[0.4489]]]),
 D2 z_norm: tensor([[[ 0.2941,  1.3503, -1.7518, -0.0350,  0.1424]],

        [[-1.0658,  1.1757,  0.9290,  0.2151, -1.2540]]])
tensor(-0.1918)