You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I have a custom dataset:
Graphs: 79
Classes: 12
Nodes in each graph : min: 4 max: 770
The task is to classify 79 graphs into 12 classes.
The parameters are to be set for node2vec: no. of walks, walk length , other arguments to be passed in the code.
How are these decided and set for such a data. Please help. I want to make histogram 2D images from this data and run a 2D CNN model.
The text was updated successfully, but these errors were encountered:
Hi, sorry for the very late reply. The parameters indeed seem to have a big impact on the results and it is not clear what are good priors for them. It obviously depends on the graphs.
Some rules of thumb that I came up with (not sure if they are the best, but at least they worked in my applications):
for networks that are quite dense and have rather low diameters, use short walk lengths (e.g., 8) and small windows (e.g., k=2 - k being half the window size)
the fewer the nodes in the networks, the smaller the embedding space should be. For instance, for less than 100 nodes, it does not make sense to have embedding spaces with more than 10 dimensions.
don't forget to add the -w flag to your calls if your networks are weighted
I have a custom dataset:
Graphs: 79
Classes: 12
Nodes in each graph : min: 4 max: 770
The task is to classify 79 graphs into 12 classes.
The parameters are to be set for node2vec: no. of walks, walk length , other arguments to be passed in the code.
How are these decided and set for such a data. Please help. I want to make histogram 2D images from this data and run a 2D CNN model.
The text was updated successfully, but these errors were encountered: