Get error when implementing custom model


Hi, sorry to bother you again.

I want to implement the model (sub module of a big one) as the image showed below.

When I run the model below, I got a error said RuntimeError: The size of tensor a (100) must match the size of tensor b (4) at non-singleton dimension 1.

It seems there are bugs in my code. Please give me some suggestions.


N,D,K= 100,4,2
def model():
    A = pyro.param('A',torch.rand(D,K))

    with pyro.plate('sample_dim',N,) as idxN:
            with pyro.plate('components',K,) as idx:

        print(z.shape) #torch.Size([2, 100])
        print(A.shape) #torch.Size([4, 2])
        temp = torch.matmul(A,z) 
        print(temp.shape) #torch.Size([4, 100])

        with pyro.plate('sample_dim_1',N,) as idxN:
            x = pyro.sample('x',dist.Normal(temp[:,idxN],t.diag(t.ones(D))))


pyro.plate('sample_dim_1',N,) as idxN will give you a tensor of indices of size N. so temp[:, idxN] has shape (4, 100) and t.diag(t.ones(D)) has shape (4,4)



It’s a stupid bug…