Can we use EvalNE for multiplex embedding ? about evalne HOT 3 CLOSED

Hello,

The current version of the library can be used to evaluate heterogeneous networks since the nature of nodes and edges is irrelevant for our LP evaluation pipeline. In fact, one of the datasets used for testing the library is heterogeneous with nodes representing students, professors, courses, etc and the edges binary relations e.g student in course, prof teaches course, etc.

About the evaluation of multiplex networks, on the other hand, I can not guarantee that the library will work without some changes. Particularly because I assume the multiplex networks you refer to will contain multiple edges between the same pairs of nodes and this could cause some of the preprocessing functions as well as the train test split algorithms to fail.

Our future plans for extending the library currently do not include multiplex network evaluation, but if you want to implement this functionality we would be very happy to include your changes :)

Alex

from evalne.

Hello Alexander.
Thank you for the answer. I am currently trying to apply EvalNE on heterogeneous networks. It is not clear for me what is the format needed for heterogeneous networks. Most of the heterogeneous networks I found like Aminer have several files to describe it and not one edgelist file.
Is it studentDB the dataset used for testing the library is heterogeneous with nodes representing students, professors, courses, etc ?
Best.

from evalne.

Hello,
The StundetDB dataset I tested the library on bundles in a single network: nodes representing professors, students, tracks etc. and edges representing the relations between these entities. If you can not generate a network like this form your data, or you want to treat specific relations differently, you can use the building blocks provided by the library and implement your own evaluation process.
Alex

from evalne.