Doesn't link prediction code need sigmoid function? #9306
Replies: 2 comments 2 replies
-
Maybe you can check to see if the representation of the model output is using a function like log_softmax() and just output the representation, because this causes the overall value of the representation to become larger. |
Beta Was this translation helpful? Give feedback.
-
it's a pleasure, in my case, I change the code from
to
This avoids situations where the value of z is too large and result of dot product is very large when calculating the probability of two points connection. You can check your forward function for similar operations that make the node representation larger |
Beta Was this translation helpful? Give feedback.
-
Hi, I have a question about the
link_pred
code in the example directory.When I calculate the probability between nodes, my code returns very large value and need to apply the sigmoid function.
But in the
link_pred
code, there was no sigmoid function for calculating the probabilities.Can anyone explain if it is a normal phenomenon returning a very large range of values after calculating the edge probability?
Beta Was this translation helpful? Give feedback.
All reactions