Bug in prediction code?

[danielegrattarola]

Hello,

I am looking at the VGRNN_prediction.py script and I found what seems to be a bug on this line:

nll_loss += self._nll_bernoulli(dec_t_sl, adj_orig_dense_list[t])

Shouldn't it be:

nll_loss += self._nll_bernoulli(dec_t_sl, adj_orig_dense_list[t + 1])

in order to do link prediction for the graph at time T+1? Did I misunderstand something?

Thanks

[Frankie123421]

I think it's actually the log-likelihood term of the loss function (equation 7), which means reconstructing A[t] by Z[t] at each time step.