Open cristianmtr opened 5 years ago
i want this too,hope somebody can give a demo
or if someone can show where the attention is in the source code model/cell,i just find in model.py,the anatation are asa fllow in line232: The weight alpha[i, j] for each hj is computed as follows: energy = a(s(i-1), H(j)) alpha = softmax(energy) Where a is a feed forward network. and in cell.py line 90,there shows how alpha calculate
anybody help?~
Hey
Is there any way to get the attention weights? That would be super nice :)