On 12.10.2016 11:59, isukritgupta wrote:
I calculate a normalisation constant for each iteration of the network generated by 'mcmc_equilibrate'.
mcmc_equilibrate() will generate fits of the SBM, i.e. partitions of the network according to the posterior probability. It will not generate networks.
This constant is the sum of probabilities of each edge existing between all pairs of nodes in the graph.
No, you should consider only all pairs of nodes that are _not_ connected in the graph. This is about predicting _missing_ edges.
To get the actual value of an edge existing in the graph, I divide the probability of that edge existing in the graph by the above mentioned constant.
Yes. (Note that this probability will correspond to the question: Assuming there is only one missing edge in the network, what is the likelihood of it being a particular edge? You have to be sure this is what you want to compute.) -- Tiago de Paula Peixoto <tiago@skewed.de>