On 22.04.2016 15:45, Andrea Briega wrote:

I have a bipartite network, with two types of nodes with no connections between the nodes of the same type, only between different types. I have run minimize_blockmodel_dl y minimize_nested_blockmodel_dl with different options but I always get separated blocks for each type of node: some blocks only with nodes of one type and the rest only with nodes of the other type. But I am really interested in the relation between this two groups of blocks. Any idea about how could I get that?

What these functions do is to fit a generative model --- the stochastic block model --- to your data. This model divides the network into equivalence classes, such that nodes in the same class have similar connections to the rest of the network. That is why it puts nodes of both partitions of a bipartite network into different groups, i.e. they are not seminar with respect to their connections. The fact that the model is capable of recognizing these types of equivalence classes is considered a feature, since it reproduces what in fact you have in your data. So, I'm not sure exactly what you want to achieve. If an algorithm places the nodes of the different partitions in the same group, it will be hiding this information from you. What else will it be hiding? How do you define what should be found, and what shouldn't? In order to do this correctly, you would need to formalize better what patterns you are searching for, and encode them in a different generative model, write an inference algorithm, etc, Now, note that the stochastic blockmodel _does_ also give you the relationship between the partitions, via the connections between them. So maybe, you can clarify a little bit more what precisely you are searching for? Best, Tiago -- Tiago de Paula Peixoto <tiago@skewed.de>