Hi Dr. Peixoto,

The attached is a complete example with the input simulated data and the python scripts. The main code is shown below as well, it's mainly one "load edges to graph" function and "minimize_nested_blockmodel_dl". Thank you!

import os

import numpy as np

import pandas as pd

import pickle

import warnings

from collections import defaultdict

import graph_tool.all as gt

def load_edges_to_graph(tidy_edges,verbose = True):

# Grab lists of unique nodes for both types

node_a_names = tidy_edges['node_a'].unique()

n_node_a = len(node_a_names)

node_b_names = tidy_edges['node_b'].unique()

n_node_b = len(node_b_names)

# create a graph-tool graph

g = gt.Graph(directed=False)

# Setup and pull out node property vectors for easy setting

name = g.vp["name"] = g.new_vp("string")

kind = g.vp["kind"] = g.new_vp("int")

# Dictionaries for adding and grabbing nodes from either class

node_a_add = defaultdict(lambda: g.add_vertex())

node_b_add = defaultdict(lambda: g.add_vertex())

num_edges, _ = tidy_edges.shape

# Start by adding all unique node_a's to make sure nodes are not jumbled by type

if(verbose): print(f'Adding all unique node As to model nodes')

for node_a_name in node_a_names:

node_a = node_a_add[node_a_name]

name[node_a] = node_a_name

kind[node_a] = 0

# Next do node b's

if(verbose): print(f'Adding all unique node Bs to model nodes')

for node_b_name in node_b_names:

node_b = node_b_add[node_b_name]

name[node_b] = node_b_name

kind[node_b] = 1

# Now go edge by edge and add the model.

five_percent = np.ceil(num_edges/20)

for edge_num in range(num_edges):

if((edge_num%five_percent == 0) and verbose):

print(f'{np.round((edge_num/num_edges)*100, decimals=1)}% done | {edge_num}/{num_edges}')

node_a_name, node_b_name = tidy_edges.iloc[edge_num,:]

# Grab the nodes for each end of the edge.

node_a = node_a_add[node_a_name]

node_b = node_b_add[node_b_name]

# Add the edge to the model

edge = g.add_edge(node_a, node_b)

return g

# Load network edges (aka phenome vectors)

tidy_edges = pd.read_csv("bisbm_sim_edge_complex.csv", dtype = 'str')

g = load_edges_to_graph(tidy_edges, verbose = True)

node_types = g.vp['kind']

niter = 1

beta = 1

# mcmc

state = gt.minimize_nested_blockmodel_dl(

state_args = dict(deg_corr = True,clabel=node_types),

multilevel_mcmc_args = dict(niter=niter,beta=beta))

Best,

Siwei

Am 20.07.22 um 08:02 schrieb Siwei Zhang:
> Hi Dr. Peixoto,
>
> I am using graph-tool version 2.45 and I have two questions.
>
> 1. I am trying to reproduce the script in the document
> 2. g = gt.collection.data["celegansneural"]
>     state = gt.minimize_nested_blockmodel_dl(g,
>     state_args=dict(overlap=True))
>
> and have the error:
> /usr/lib/python3/dist-packages/graph_tool/inference/blockmodel.py:390:
> UserWarning: unrecognized keyword arguments: ['overlap']
> warnings.warn("unrecognized keyword arguments: " +
>
> It seems the argument of "overlap" is removed.

The proper way to use an overlapping model is to pass the option:

     state_args=dict(base_type=OverlapBlockState)

> 2. Regardless of the question1, I am trying to do a bipartite
> version stochastic block model and I define "clabel" to constraint
> labels on the vertices so that vertices with different label values will
> not be clustered in the same group. But I always have the error of
> "ValueError: cannot move vertex across clabel barriers". The below is
> the code:
> node_types = g.vp['kind']
> node_types.get_array()
> Output: PropertyArray([1, 1, 1, ..., 2, 2, 2], dtype=int32)
>
> state = gt.minimize_nested_blockmodel_dl(
> g,
> state_args=dict(clabel=node_types,pclabel=node_types,deg_corr=True),
> multilevel_mcmc_args = dict(niter=niter,beta=beta))
> Could you please help me with these questions? Thanks!
In order to understand what is happening you would need to send us a
minimal but complete working example that shows the problem.

Best,
Tiago

--
Tiago de Paula Peixoto <tiago@skewed.de>