Weighted SBMs: PPBlockState, Model Class Selection

[Srinivas_Govinda_Sur (Srinivas Govinda Surampudi)]

Hi Prof. Peixoto,

gt version: 2.77
OS: ubuntu 22

I am working with the weighted SBMs. I have two questions:

1. Does PPBlockState work with edge weights? I could not find how to pass recs and rec_types arguments to the PPBlockState. I am able to pass those arguments to BlockState, and NestedBlockState.
2. Model Class Selection: in the paper peixoto-revealing-2021 and in the section of the cookbook model class selection, does the procedure work for weighted SBMs or is it only for simple SBMs? I believe it does, because we are able to sample partitions using the same mcmc_equilibrate function.

Many thanks,
Govinda

[tiago (Tiago Peixoto)]

@Srinivas_Govinda_Sur

1. Does PPBlockState work with edge weights? I could not find how to pass recs and rec_types arguments to the PPBlockState. I am able to pass those arguments to BlockState, and NestedBlockState.

No, this is not implemented yet!

@Srinivas_Govinda_Sur

Model Class Selection: in the paper peixoto-revealing-2021 and in the section of the cookbook model class selection, does the procedure work for weighted SBMs or is it only for simple SBMs? I believe it does, because we are able to sample partitions using the same mcmc_equilibrate function.

Yes, that is applicable to those models as well.

[Srinivas_Govinda_Sur (Srinivas Govinda Surampudi)]

Thank you!
So just so I understand correctly:
when I collect the description lengths of the sampled partitions with mcmc_equilibrate(), I will need to take into account the scaling of the probability density incurred by the variable transformations. I can then use the modified description lengths in calculating the total evidence and select the model class.