New hidden vector computation in GNNCell of PyTorch code about sr-gnn HOT 3 CLOSED

Yes. n_node includes the padding item.

from sr-gnn.

Q1:

Thank you for pointing out. The code and the paper are not matched as for this equation. I referred to this code when I wrote this part but did not notice this mistake and the fact that their code and paper are not matched either.

Q2:

We use the 0th item to pad the sequences with different lengths. So when we try to generate recommendations, we need to exclude the 0th item out of the potential targets.

from sr-gnn.

Q1

I tested with the equation form as Eq(5) with hy = hidden + inputgate * (newgate - hidden). The result becomes worse. It is not accord with intuition.

Q2

So in the code,

    if opt.dataset == 'diginetica':
        n_node = 43098
    elif opt.dataset == 'yoochoose1_64' or opt.dataset == 'yoochoose1_4':
        n_node = 37484
    else:
        n_node = 310

The actual number of item = n_node - 1?

from sr-gnn.