I couldn’t log in to Gitlab for some reason, so I decided to post this here.
I have a few questions regarding graph-tool. The first two questions are about potential bugs.
In graph-tool (version 2.58), it doesn’t allow me to use PPBlockState as a base_type for the lowermost level of NestedBlockState. It didn’t work in version 2.55 either. It returns the following error:
TypeError: PPBlockState.__init__() got an unexpected keyword argument 'Lrecdx'
Is this a bug or just a missing implementation?
In version 2.55, B_max worked correctly, but in version 2.58, when provided for multilevel_mcmc_sweep in minimize_nested_blockmodel_dl(), it doesn’t follow the B_max value. It returns partitions with a larger number of blocks than the predefined B_max. Bug?
It is very unclear to me how to provide the initial partition for bs in NestedBlockState. I have tried multiple approaches, but all of them fail to return the correct summary (print_summary()). I already have a partition, and I just want to compute the entropy for it. My approach was the following:
Assume I have a graph with 1000 nodes. On the first level, I have four blocks. On the second level, I have two blocks, and finally, on the third level, one block containing all the nodes.
I created a VertexPropertyMap for each level that contains 1000 elements. For each level, there is a VPM that indicates the membership of all the 1000 nodes. Obviously, the number of unique elements is, for the first level, 4; for the second level, 2; and for the third level, 1. This doesn’t seem to be the right approach, as the print_summary() returns the wrong number of blocks for each level. How am I supposed to “encode” my initial partition correctly?