Confusion about your DeepWalk implementation about graph-learn

[TheFinalHydra]

Never mind, I think I did misunderstand it. I get it now, that each array in the path is the collection node ids in each degree.

[TheFinalHydra]

But what if the graph it too sparse that you cannot find a path of length walk_len? Will graph-learn pad the path of just discard it?

[baoleai]

In fact, for undirected graphs, you can always sample a path of length walk-len .

[TheFinalHydra]

In fact, for undirected graphs, you can always sample a path of length walk-len .

Yes I get it, thanks for pointing out, but for some directed graphs, how do you solve such potential issue? Is it a good idea to separately apply SkipGram on 2-degree paths, 3-degree paths, etc... or we should avoid using DeepWalk but other algorithms on such sparse directed graph?

[baoleai]

I think you can modify the implementation of deepwalk, propose your solution to this situation, welcome to contribute:)