[graph-tool] Getting edge probabilities for SBM.

Tiago de Paula Peixoto tiago at skewed.de
Wed Oct 12 11:34:33 CEST 2016


On 12.10.2016 07:12, isukritgupta wrote:
> Hi Tiago, I am a new user for Graph Tool.  I was trying to do a fitting
> approach for a graph. And to measure the goodness of fit, I intend to use
> log of likelihood score for the generated graph and the real graph.
> 
> For this, I needed to calculate the probabilities of each edge existing in
> the generated graph. However, when I use the 'get_edge_prob' function for
> BlockState, I get values which are greater than 0 for an edge existing.
> Which should not be possible since log(prob) <= 0. It is supposed to return
> 'unnormalised log probability' of the edge.
> 
> Can you please clarify?

The exact quantity computed is explained here:

https://graph-tool.skewed.de/static/doc/demos/inference/inference.html#predicting-spurious-and-missing-edges

It is the posterior likelihood of a missing edge, conditioned on the
observed graph and model. It is unnormalized, because computing the
normalization constant would require obtaining this value for every possible
missing edge, which is typically of the order O(N^2).

Since it is not normalized, it can also return log-probability values that
are above zero.

Best,
Tiago


-- 
Tiago de Paula Peixoto <tiago at skewed.de>

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 801 bytes
Desc: OpenPGP digital signature
URL: <https://lists.skewed.de/pipermail/graph-tool/attachments/20161012/5517a405/attachment.asc>


More information about the graph-tool mailing list