Hi Prof. Peixoto,
I’m running into an alignment issue when using PartitionModeState() in graph-tool;
v2.92 on Ubuntu 22, Python 3.9.
I’m attempting to align an ensemble of partitions that consists of two pre-aligned subsets (i.e., partitions within each subset are already label-consistent). However, PartitionModeState does not seem to resolve the alignment across the two subsets, no matter the variation of arguments I try.
MWE:
import numpy as np
import dill as pickle
import graph_tool.all as gt
# load partitions
with open('bs_1.pkl', 'rb') as f:
[bs1] = pickle.load(f)
print(f'len of bs1: {len(bs1)}')
with open('bs_2.pkl', 'rb') as f:
[bs2] = pickle.load(f)
print(f'len of bs2: {len(bs2)}')
bs = bs1 + bs2
# align partitions
pmode = gt.PartitionModeState(bs, relabel=True, nested=False, converge=False)
delta = 1
while np.abs(delta) > 1e-8:
delta = pmode.replace_partitions()
print(delta)
bs_aligned = [v for k, v in sorted(pmode.get_partitions().items())]
# Alignment checks
# 1. Within bs1
x, y = bs_aligned[23], bs_aligned[0]
print(gt.align_partition_labels(x, y) - x)
# 2. Within bs2
x, y = bs_aligned[89], bs_aligned[67]
print(gt.align_partition_labels(x, y) - x)
# 3. Across bs1 and bs2
x, y = bs_aligned[89], bs_aligned[0]
print(gt.align_partition_labels(x, y) - x)
I check whether the partitions are aligned using the gt.align_partition_labels(x, y) function. x is from bs1 , and y from bs2 . If the partitions are aligned across the sets, then this function should return the same x.
As expected, partitions within each set remain aligned. However, partitions across the two sets remain unaligned, even after convergence of pmode.
Is there anything additional I should incorporate to force PartitionModeState to resolve alignment across these subgroups?