Description length of consensus partition

graph-tool
1

Dear Tiago, all,

I'm happy to have discovered the new partition overlap functionality. It works very nicely. But is it possible to obtain the description length of a nested consensus state?

Getting the description length of the consensus state works fine for an unnested model:

from graph_tool.all import *
h = collection.data['polbooks']
num_models = 2
states = [minimize_blockmodel_dl(h) for i in range(num_models)]
bs = [states[i].get_blocks().a for i in range(num_models)]
c, r = partition_overlap_center(bs)

h_blocks = h.new_vp('int')
h_blocks.a = c
BlockState(h, h_blocks).entropy()

For nested partitions, getting the consensus state is a bit more complicated:

states_nested = [minimize_nested_blockmodel_dl(h) for i in range(num_models)]
bs_nested = []
for j in range(num_models):
    l = []
    for k in range(len(states_nested[j].get_levels())):
        l.append(states_nested[j].get_bstack()[k].vp.b.a)
    bs_nested.append(l)
c_nested, r_nested = nested_partition_overlap_center(bs_nested)

But how can I get its description length?

Many thanks for your attention

Haiko
GESIS, Cologne

attachment.html (3.71 KB)

2

Ni! Hi Haiko,

Do you mean the nested description length as in:

NestedBlockState(h, c_nested).entropy()

?

.~´

attachment.html (3.49 KB)

3

Do you mean the nested description length as in:

NestedBlockState(h, c_nested).entropy()

Oh man, of course, NestedBlockState() is what I was looking for!

Thx, H

attachment.html (1.81 KB)