Nested partition overlap

Dear Tiago,

I just saw your talk at the Networks 2021 satellite symposium on "Community Structure in Networks." Congratulations on presenting a complicated paper (https://journals.aps.org/prx/abstract/10.1103/PhysRevX.11.021003) in an accessible way. In that paper, you develop ways to return more meaningful (i.e., based on consensus) point estimates from the posterior distribution of the SBM. In the talk, you said the method can just as well be applied to multiple solutions of any clustering algorithm.

The partition_overlap_center() function for flat partitions works fine with either lists from a posterior distribution or from multiple runs of an SBM. But the nested_partition_overlap_center() function crashes (version 2.37) when the nested partitions (for which the consensus is to be obtained) do not have the same depth.

Is there a principled problem with making nested_partition_overlap_center() applicable to nested partitions of different depth? Is it something you would like to do for the next release or is there a quick fix?

Best wishes

Haiko

attachment.html (3.06 KB)

Dear Haiko,

Dear Tiago,

I just saw your talk at the Networks 2021 satellite symposium on
“Community Structure in Networks.” Congratulations on presenting a
complicated paper
(https://journals.aps.org/prx/abstract/10.1103/PhysRevX.11.021003
<https://journals.aps.org/prx/abstract/10.1103/PhysRevX.11.021003&gt;\) in
an accessible way. In that paper, you develop ways to return more
meaningful (i.e., based on consensus) point estimates from the posterior
distribution of the SBM. In the talk, you said the method can just as
well be applied to multiple solutions of any clustering algorithm.

The partition_overlap_center() function for flat partitions works fine
with either lists from a posterior distribution or from multiple runs of
an SBM. But the nested_partition_overlap_center() function crashes
(version 2.37) when the nested partitions (for which the consensus is to
be obtained) do not have the same depth.

Please add an issue in the website with a minimal working example that
shows the problem, and it will be fixed.

Is there a principled problem with making
nested_partition_overlap_center() applicable to nested partitions of
different depth? Is it something you would like to do for the next
release or is there a quick fix?

This trivial to fix by adding levels to the shortest hierarchy with a
single group each. I probably just forgot to account for this, since it
didn't occur to me to pass it two hierarchies of different depths.

Best,
Tiago