torch.bmm(a,b)

函数作用:实现batch的矩阵乘法
计算两个tensor的矩阵乘法，torch.bmm(a,b),tensor a 的size为(b,h,w),tensor b的size为(b,w,h),注意两个tensor的维度必须为3.
代码示例

>>> c=torch.randn((2,5))
>>> print(c)
tensor([[ 1.0559, -0.3533,  0.5194,  0.9526, -0.2483],
        [-0.1293,  0.4809, -0.5268, -0.3673,  0.0666]])
>>> d=torch.reshape(c,(5,2))
>>> print(d)
tensor([[ 1.0559, -0.3533],
        [ 0.5194,  0.9526],
        [-0.2483, -0.1293],
        [ 0.4809, -0.5268],
        [-0.3673,  0.0666]])
>>> e=torch.bmm(c,d)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
RuntimeError: Dimension out of range (expected to be in range of [-2, 1], but got 2)

当tensor维度为2时会报错！

>>> ccc=torch.randn((1,2,2,5))
>>> ddd=torch.randn((1,2,5,2))
>>> e=torch.bmm(ccc,ddd)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
RuntimeError: invalid argument 1: expected 3D tensor, got 4D at /opt/conda/conda-bld/pytorch_1535490206202/work/aten/src/TH/generic/THTensorMath.cpp:2304

维度为4时也会报错！

>>> cc=torch.randn((2,2,5))
>>>print(cc)
tensor([[[ 1.4873, -0.7482, -0.6734, -0.9682,  1.2869],
         [ 0.0550, -0.4461, -0.1102, -0.0797, -0.8349]],
         [[-0.6872,  1.1920, -0.9732,  0.4580,  0.7901],
         [ 0.3035,  0.2022,  0.8815,  0.9982, -1.1892]]])
>>>dd=torch.reshape(cc,(2,5,2))
>>> print(dd)
tensor([[[ 1.4873, -0.7482],
         [-0.6734, -0.9682],
         [ 1.2869,  0.0550],
         [-0.4461, -0.1102],
         [-0.0797, -0.8349]],
         [[-0.6872,  1.1920],
         [-0.9732,  0.4580],
         [ 0.7901,  0.3035],
         [ 0.2022,  0.8815],
         [ 0.9982, -1.1892]]])
>>>e=torch.bmm(cc,dd)
>>> print(e)
tensor([[[ 2.1787, -1.3931],
         [ 0.3425,  1.0906]],
         [[-0.5754, -1.1045],
         [-0.6941,  3.0161]]])
 >>> e.size()
torch.Size([2, 2, 2])
正确！
————————————————
版权声明：本文为CSDN博主「wang xiang」的原创文章，遵循CC 4.0 BY-SA版权协议，转载请附上原文出处链接及本声明。
原文链接：https://blog.csdn.net/qq_40178291/article/details/100302375