remap_ids significantly slows down for weights with a large number of neighbors

Platform information: posix linux posix.uname_result(sysname='Linux', nodename='barguzin-tp', release='5.11.0-34-generic', version='#36~20.04.1-Ubuntu SMP Fri Aug 27 08:06:32 UTC 2021', machine='x86_64')
Python version: 3.9.2 | packaged by conda-forge | (default, Feb 21 2021, 05:02:46)
SciPy version: 1.7.0
NumPy version: 1.20.2

I am working with CA census tracts, but any shapefile with >2000 rows would suffice. The problem is with libpysal.weight.util.block_weights function, which runs infinitely after a certain threshold of dataframe size is met. I tried running the function on my laptop at first (see specs above). My first guess was that it could be related to the amount of RAM or CPU usage, so I tried running it on the server (256GB RAM, 32 CPU cores, CentOS). In both cases the computer seems to go into this infinite computation loop at the size of 1500-1600 rows.

Here is the reproducible code demo:

wget 'https://www2.census.gov/geo/tiger/TIGER2019/TRACT/tl_2019_06_tract.zip'; unzip tl_2019_06_tract.zip -d tracts

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt 
import geopandas as gpd 
import libpysal
import os

# define function that generates random numbers 
def get_random_classes(n, var, dataframe):
    '''
    generate random variable containing n-classes and create a new variable in the dataframe
    '''
    dataframe["{}".format(var)] = np.random.randint(n, size=dataframe.shape[0])

tracts = gpd.read_file('tracts/tl_2019_06_tract.shp')
print(tracts.shape)

get_random_classes(n = 3, var = 'clusters', dataframe = tracts)

# create spatial weights matrix
# W = libpysal.weights.Queen.from_dataframe(tracts)

sam1 = tracts.sample(n=2000) # change sample size here

w_bl_sam1 = libpysal.weights.util.block_weights(
    sam1['clusters'].values, 
    ids=sam1.index
)

pysal / libpysal

remap_ids significantly slows down for weights with a large number of neighbors #433