3D tensor * 2D tensor dot in Torch

Question

In Theano, when I have a 3D tensor x with shape [A,B,C] and a 2D tensor y with shape [C,D], then theano.tensor.dot(x, y) returns a 3D tensor with shape [A,B,D].

What would be the equivalent operation in Torch? torch.dot doesn't seem to do that, and x * y and torch.mm complain that they want a 2D tensor for both arguments, and torch.bmm wants 3D tensors for both arguments.

Answer 1

As @smhx proposed, the possible solution is to repeat the second tensor (there's a way to do it without memory allocating) and then perform a batch matrix matrix product:

function repeatNoCopy(tensor, k)
    local tens_size = tensor:size():totable()
    return torch.expand(tensor:view(1, unpack(tens_size)), k, unpack(tens_size))
end

A = torch.rand(3, 2, 5)
B = torch.rand(5, 4)
B_rep = repeatNoCopy(B, 3)

result = torch.bmm(A, B_rep)

print(result)
> [torch.DoubleTensor of size 3x2x4]