Fitting SBM with different priors

graph-tool
1

Dear Tiago,

I would like to fit an SBM with the /minimize_blockmodel_dl()/ function.
Specifically, I would like to customize the optimization procedure with
different priors for the model parameters. I am aware that
/BlockState.entropy()/ returns the entropy (for fitting to SBM) with
*labelled* input (partition & degree sequence), and /model_entropy()/
returns the entropy (for constructing the model) with *static* input (B, N,
E). However, I don't see an argument in the /minimize_blockmodel_dl()/
function that I could enforce certain parameter priors at the first place,
be it /degree_dl_kind == "uniform"/ or /degree_dl_kind == "distributed"/.

Do I miss something from the documentation? For example, may I customize
/state_args/ in /minimize_blockmodel_dl()/ for this purpose?

Sincerely thanks,
Tzu-Chi

2

The function minimize_blockmodel_dl() calls many other functions which need
to compute the entropy (among other things), so things are organized in a way
to make the code simpler, and contain the explosion of function parameters,
but it makes options for customization like this a bit hidden. To achieve
what you want, you need to do:

    minimize_blockmodel_dl(g, mcmc_args=dict(entropy_args=dict(degree_dl_kind="uniform")))

Best,
Tiago