tensor乘运算

torch.mul(a, b) 是矩阵 对应位相乘，即点乘操作， a和b的维度必须相等，a的维度是(1,2)， 则b的维度必须是（1,2）， 返回还是（1,2）的矩阵

torch.mm(a,b)是矩阵a和b矩阵相乘，a的维度是（1,2），b的维度是（2,3），返回是（1,3）的矩阵

torch.bmm(a,b)是矩阵a和b在维度1、2上矩阵相乘，一般要求是 三维矩阵，a的维度是（64,1,2），b的维度是（64,2,3）返回的是（64,1,3）矩阵

import torch


if __name__ == "__main__":
    x = torch.ones(1, 2)
    y = torch.ones(1, 2) * 2
    z = torch.mul(x, y)
    print("torch.mul() example ")
    print("x.shape is ", x.shape) 
    print("y.shape is ", y.shape)
    print("z.shape is ", z.shape) ## [1, 2]

    x = torch.ones(1, 2)
    y = torch.ones(2, 3) 
    z = torch.mm(x, y)
    print("torch.mm() example ")
    print("x.shape is ", x.shape)
    print("y.shape is ", y.shape)
    print("z.shape is ", z.shape) ## [1,3]

    x = torch.randn(64, 1, 2)
    y = torch.randn(64, 2, 3)
    z = torch.bmm(x, y)
    print("torch.bmm example ")
    print("x.shape is ", x.shape)
    print("y.shape is ", y.shape)
    print("z.shape is ", z.shape) ## [64, 1, 3]

转载于:https://www.cnblogs.com/yeran/p/11288171.html