software for identifying modules in networks (e.g. "community detection"), as described in "a bayesian approach to network modularity" (physical review letters 100, 258701 (2008); http://link.aps.org/abstract/PRL/v100/e258701).
Be the first to post a text review of variational bayes for network modularity. Rate and review a project by clicking thumbs up or thumbs down in the right column.
* changed order of updates to "E step", then "M step", then evaluation of maximized free energy. variational hyperparameters are initialized to prior hyperparameters * changed evaluation of free energy to come after VBM step, as is assumed in calculations * updated VBE-step to do per-row updates in mex file (for loop over nodes is much faster in c than matlab) * included matlab function (at bottom of file) as backup for those who can't compile mex * changed lnpi calculation to be evaluated in psi( ) and *then* replicated, reducing number of calls to psi( ) by factor of N * changed repmats to use index trick for replicating rows/columns, as in http://www.psi.toronto.edu/~vincent/matlabindexrepmat.html * added display of F for each iteration if VERBOSE>=2 * turned on warning for increase in F by default * initial Q is saved to net.Q0 if opts.SAVE_ITER=1; * saves hyperparameters for prior if SAVE_ITER==1 * fixed small typo in storing variables during saving of iterations * Q0 is now initialized as a uniform random stochastic matrix * added rndx=rndx(1) in the event that several restarts converge to the same minimum free energy value
. changed order of updates to "E step", then "M step", then evaluation of maximized free energy. variational hyperparameters are initialized to prior hyperparameters and the initial Q is saved to net.Q0 if opts.SAVE_ITER=1; . Q0 is now initialized as a uniform random stochastic matrix, as opposed to "flat" with some hard-set nodes
. changed evaluation of free energy to come after VBM step, as is assumed in calculations; this doesn't change updates, just evaluation of objective function . note: updates on Q should really be one row at a time and subtracting a "diagonal" term. todo: add this as an option in future versions.
Be the first person to add a text review.
Copyright © 2009 Geeknet, Inc. All rights reserved. Terms of Use
Thanks for your rating!
Would you also like to write a review?