Closed BradKML closed 3 years ago
I apologize, but we do not understand what do you wish. What do you mean by "Allow the ... to be made into a module"? What do you mean by "Selector"? Where would the outputs that you describe be used?
Does this refer to our project and how?
Note that Orange doesn't include anything related to networks. There is an add-on, but it no longer uses networkx.
There is an add-on, but it no longer uses networkx That is unexpected that Orange3 is not as tightly tied to Network Node Classification tools.
The add-on used networkx until we realized that the only thing that we used were
Hence we stopped using networkx and instead implemented the necessary data structures and code ourselves. Some statistics are still missing, but those didn't properly work in networkx, neither.
But I still don't understand what do you actually wish. Please read the first two paragraphs of my previous comment again.
It relies on the Graph objects of NetworkX and iGraph to work. Ref:
KarateClub is another library that provides the feature of "Node Embedding" derived from NetworkX objects
There are examples that apply this into Node Role Classification (e.g. Twitter KOL/follower distinction)
https://github.com/benedekrozemberczki/RolX and https://github.com/dkaslovsky/GraphRole
[x] What's your proposed solution?
Allow the use of evaluation metrics to be made into a module Input: Predicted node partition, Ground Truth data Selector: Accuracy measures e.g. NMI, ARI, Purity... Output: Accuracy Index
Allow the centrality measure to be made into a module Input: NetworkX Graph Selector: Centrality measurements Output: Dictionary of node names as keys and centrality quantification as values