[graph-tool] Re: Model specification and extraction of layer-specific partitions in LayeredBlockState when overlap=True

Am 24.08.21 um 12:49 schrieb arttu.malkamaki@helsinki.fi:

What is the difference between A) fitting a separate model to each layer and B) fitting an overlapping model with layers=True? When interpreting my results, what can I say about the dependencies between layers with discrete types of relationships?

With B) the overlapping partition is encoded in a manner that takes correlations between the layers into account, so that the group memberships will vary between layers only if there is enough evidence requiring this. With A) the independent models do not take advantage of any similarity between layers.