Relative thresholding but retain edges to keep graph connected

Submitted by minihk on

Dear all,

We would like to apply relative thresholding (i.e. density thresholding) to our correlation matrices. This can be easily done using graphvar or BCT toolbox (threshold_proportional.m function), but we would like to keep connections in the graph if by removing them it would cause the graph to disconnect. I.e. recursively removing edges, starting with the weakest weights and progressing until the density threshold (e.g. 20%) is reached and within this iterative process retain any edge that would cause the graph to disconnect by its removal, even in the case of a low weight. Are there any matlab scripts available for this within the graphvar or BCT toolbox (as I am not expert enough to create such a script myself)?

Best,

Mini

 

sandywang

Tue, 11/17/2015 - 16:20

Permalink

I think you should create a yourself script to do it Maybe using MatlabBGL's function components which is used to get the number and index of all components in a network and help you judge whether it is a connected network. :) Best, Sandy

Johann Kruschwitz

Tue, 11/17/2015 - 16:35

Permalink

... alternatively you could also use the FragCheck function implemented in GraphVar beta 0.62 that will show you at which threshold network fragmentation occurs for each subject. Based on these results (every subject will have it´s own personal threshold) you would know which tresholds to use per subect to construct a fully connected graph with minimum density.

I hope this is what you are looking for,
Johann

Forums
GraphVar