I tried this operation for one batch and worked.
x = torch.FloatTensor([[ax,bx],[cx,dx],[ex,fx],[gx,hx]])
y = torch.FloatTensor([[ay,by],[cy,dy],[ey,fy],[gy,hy]])
z = x[:,None,0] * y[:,1]
print(z)
tensor([[ ax*by, ax*dy, ax*fy, ax*hy],
[ cx*by, cx*dy, cx*fy, cx*hy],
[ ex*by, ex*dy, ex*fy, ex*hy],
[ gx*by, gx*dy, gx*fy, gx*hy]])
I want to repeat above operation for several batches like below example without using for loop.
x = torch.FloatTensor([[[ax1,bx1],[cx1,dx1],[ex1,fx1],[gx1,hx1]],
[[ax2,bx2],[cx2,dx2],[ex2,fx2],[gx2,hx2]]])
y = torch.FloatTensor([[[ay1,by1],[cy1,dy1],[ey1,fy1],[gy1,hy1]],
[[ay2,by2],[cy2,dy2],[ey2,fy2],[gy2,hy2]]])
This is what I want to get for result.
tensor([[[ ax1*by1, ax1*dy1, ax1*fy1, ax1*hy1],
[ cx1*by1, cx1*dy1, cx1*fy1, cx1*hy1],
[ ex1*by1, ex1*dy1, ex1*fy1, ex1*hy1],
[ gx1*by1, gx1*dy1, gx1*fy1, gx1*hy1]],
[[ ax2*by2, ax2*dy2, ax2*fy2, ax2*hy2],
[ cx2*by2, cx2*dy2, cx2*fy2, cx2*hy2],
[ ex2*by2, ex2*dy2, ex2*fy2, ex2*hy2],
[ gx2*by2, gx2*dy2, gx2*fy2, gx2*hy2]]])
I tried as below, but gave me errors
z = x[...,None,0] * y[...,1]
RuntimeError: The size of tensor a (4) must match the size of tensor b (2) at non-singleton dimension 1
Is there any way doing this without for loop?