Recovering degree correction parameters from minimize_blockmodel_dl


I am trying to normalize the output of BlockState.get_matrix() based on the degree correction so that we can show the implied parameters of the Poisson distribution, ie the \lambda_{rs} in E[i \in r \to j \in s] = \lambda_{rs}\theta_i\theta_j

I am following Taigo’s paper Statistical inference of assortative community structures (, where equation 10 gives a nonlinear system to solve for the block level \theta_r.

But I realize I am not sure which flavor of SBM minimize_blockmodel_dl is implementing, and if the above paper is the right reference. Any guidance on this would be greatly appreciated.

1 Like

The model used by BlockState is the general SBM (not necessarily assortative), as is explained in the documentation. See the details in: BlockState — graph-tool 2.68 documentation

If you want the assortative version, use PPBlockState, as explained here: Inferring modular network structure — graph-tool 2.68 documentation

1 Like