Problems with the distance matrices

[pandadada123]

Thanks for this great package for FGW!

But there may be some bugs with the distance matrices C1 and C2. I tried a graph in the dataset BZR, and some entries in C1 seem to be not correct.

Here are the codes.

import numpy as np
import os,sys
sys.path.append(os.path.realpath('../lib'))

from graph import graph_colors,draw_rel,draw_transp,Graph,wl_labeling
from ot_distances import Fused_Gromov_Wasserstein_distance,Wasserstein_distance
import copy
from data_loader import load_local_data,histog,build_noisy_circular_graph
import matplotlib.pyplot as plt
import networkx as nx
import ot

dataset_n='bzr'
path='../data/'
X,label=load_local_data(path,dataset_n,wl=0) 
G1=X[0]
g1=G1.nx_graph

G2 = copy.deepcopy(G1)

vmin=-5
vmax=20 # the range of color

plt.figure(figsize=(8,5))
draw_rel(g1,vmin=vmin,vmax=vmax,with_labels=True,draw=False)

dgw=Fused_Gromov_Wasserstein_distance(alpha=1,features_metric='hamming_dist',method='shortest_path').graph_d(G1,G2)

The graph G1 I get is

I also slightly changed the following two lines to draw this graph properly.

FGW/lib/graph.py

Line 355 in 3d2128a

val_map[k]=cpick.to_rgba(v)

FGW/lib/graph.py

Line 379 in 3d2128a

nx.draw(G,pos,with_labels=True,labels=nx.get_node_attributes(G,'attr_name'),node_color = colors)

Change them to:

val_map[k]=cpick.to_rgba(sum(pow(np.array(v),2)))

and

 nx.draw_networkx(G,pos,node_color = colors)

The distance matrix (30 * 30) of this graph can be checked. The first line of it is:

0	1	1	1	2	2	3	3	2	4	3	2	5	5	5	4	6	6	5	4	6	6	7	7	8	8	7	9	8	7

where the third entry "1" indicates the shortest path length between node 1 and node 3, but we can see from the graph that this length should be "2".

I would appreciate any comments you may have!