LGGM-Text2Graph / analysis /spectre_utils.py
YuWang0103's picture
Upload 41 files
6b59850 verified
raw
history blame
36.1 kB
###############################################################################
#
# Adapted from https://github.com/lrjconan/GRAN/ which in turn is adapted from https://github.com/JiaxuanYou/graph-generation
#
###############################################################################
# import graph_tool.all as gt
##Navigate to the ./util/orca directory and compile orca.cpp
# g++ -O2 -std=c++11 -o orca orca.cpp
import os
import copy
import torch
import torch.nn as nn
import numpy as np
import networkx as nx
import subprocess as sp
import concurrent.futures
import pygsp as pg
import secrets
from string import ascii_uppercase, digits
from datetime import datetime
from scipy.linalg import eigvalsh
from scipy.stats import chi2
from analysis.dist_helper import compute_mmd, gaussian_emd, gaussian, emd, gaussian_tv, disc
from torch_geometric.utils import to_networkx
import wandb
from collections import defaultdict
PRINT_TIME = False
__all__ = ['degree_stats', 'clustering_stats', 'orbit_stats_all', 'spectral_stats', 'eval_acc_lobster_graph']
def degree_worker(G):
return np.array(nx.degree_histogram(G))
def degree_stats(graph_ref_list, graph_pred_list, is_parallel=True, compute_emd=False):
''' Compute the distance between the degree distributions of two unordered sets of graphs.
Args:
graph_ref_list, graph_target_list: two lists of networkx graphs to be evaluated
'''
sample_ref = []
sample_pred = []
# in case an empty graph is generated
graph_pred_list_remove_empty = [
G for G in graph_pred_list if not G.number_of_nodes() == 0
]
prev = datetime.now()
if is_parallel:
with concurrent.futures.ThreadPoolExecutor() as executor:
for deg_hist in executor.map(degree_worker, graph_ref_list):
sample_ref.append(deg_hist)
with concurrent.futures.ThreadPoolExecutor() as executor:
for deg_hist in executor.map(degree_worker, graph_pred_list_remove_empty):
sample_pred.append(deg_hist)
else:
for i in range(len(graph_ref_list)):
degree_temp = np.array(nx.degree_histogram(graph_ref_list[i]))
sample_ref.append(degree_temp)
for i in range(len(graph_pred_list_remove_empty)):
degree_temp = np.array(
nx.degree_histogram(graph_pred_list_remove_empty[i]))
sample_pred.append(degree_temp)
# mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=gaussian_emd)
# mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=emd)
if compute_emd:
# EMD option uses the same computation as GraphRNN, the alternative is MMD as computed by GRAN
# mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=emd)
mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=gaussian_emd)
else:
mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=gaussian_tv)
# mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=gaussian)
elapsed = datetime.now() - prev
if PRINT_TIME:
print('Time computing degree mmd: ', elapsed)
return mmd_dist
###############################################################################
def spectral_worker(G, n_eigvals=-1):
# eigs = nx.laplacian_spectrum(G)
try:
eigs = eigvalsh(nx.normalized_laplacian_matrix(G).todense())
except:
eigs = np.zeros(G.number_of_nodes())
if n_eigvals > 0:
eigs = eigs[1:n_eigvals + 1]
spectral_pmf, _ = np.histogram(eigs, bins=200, range=(-1e-5, 2), density=False)
spectral_pmf = spectral_pmf / spectral_pmf.sum()
return spectral_pmf
def get_spectral_pmf(eigs, max_eig):
spectral_pmf, _ = np.histogram(np.clip(eigs, 0, max_eig), bins=200, range=(-1e-5, max_eig), density=False)
spectral_pmf = spectral_pmf / spectral_pmf.sum()
return spectral_pmf
def eigval_stats(eig_ref_list, eig_pred_list, max_eig=20, is_parallel=True, compute_emd=False):
''' Compute the distance between the degree distributions of two unordered sets of graphs.
Args:
graph_ref_list, graph_target_list: two lists of networkx graphs to be evaluated
'''
sample_ref = []
sample_pred = []
prev = datetime.now()
if is_parallel:
with concurrent.futures.ThreadPoolExecutor() as executor:
for spectral_density in executor.map(get_spectral_pmf, eig_ref_list,
[max_eig for i in range(len(eig_ref_list))]):
sample_ref.append(spectral_density)
with concurrent.futures.ThreadPoolExecutor() as executor:
for spectral_density in executor.map(get_spectral_pmf, eig_pred_list,
[max_eig for i in range(len(eig_ref_list))]):
sample_pred.append(spectral_density)
else:
for i in range(len(eig_ref_list)):
spectral_temp = get_spectral_pmf(eig_ref_list[i])
sample_ref.append(spectral_temp)
for i in range(len(eig_pred_list)):
spectral_temp = get_spectral_pmf(eig_pred_list[i])
sample_pred.append(spectral_temp)
# mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=gaussian_emd)
if compute_emd:
mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=emd)
else:
mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=gaussian_tv)
# mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=gaussian)
elapsed = datetime.now() - prev
if PRINT_TIME:
print('Time computing eig mmd: ', elapsed)
return mmd_dist
def eigh_worker(G):
L = nx.normalized_laplacian_matrix(G).todense()
try:
eigvals, eigvecs = np.linalg.eigh(L)
except:
eigvals = np.zeros(L[0, :].shape)
eigvecs = np.zeros(L.shape)
return (eigvals, eigvecs)
def compute_list_eigh(graph_list, is_parallel=False):
eigval_list = []
eigvec_list = []
if is_parallel:
with concurrent.futures.ThreadPoolExecutor() as executor:
for e_U in executor.map(eigh_worker, graph_list):
eigval_list.append(e_U[0])
eigvec_list.append(e_U[1])
else:
for i in range(len(graph_list)):
e_U = eigh_worker(graph_list[i])
eigval_list.append(e_U[0])
eigvec_list.append(e_U[1])
return eigval_list, eigvec_list
def get_spectral_filter_worker(eigvec, eigval, filters, bound=1.4):
ges = filters.evaluate(eigval)
linop = []
for ge in ges:
linop.append(eigvec @ np.diag(ge) @ eigvec.T)
linop = np.array(linop)
norm_filt = np.sum(linop ** 2, axis=2)
hist_range = [0, bound]
hist = np.array([np.histogram(x, range=hist_range, bins=100)[0] for x in norm_filt]) # NOTE: change number of bins
return hist.flatten()
def spectral_filter_stats(eigvec_ref_list, eigval_ref_list, eigvec_pred_list, eigval_pred_list, is_parallel=False,
compute_emd=False):
''' Compute the distance between the eigvector sets.
Args:
graph_ref_list, graph_target_list: two lists of networkx graphs to be evaluated
'''
prev = datetime.now()
class DMG(object):
"""Dummy Normalized Graph"""
lmax = 2
n_filters = 12
filters = pg.filters.Abspline(DMG, n_filters)
bound = np.max(filters.evaluate(np.arange(0, 2, 0.01)))
sample_ref = []
sample_pred = []
if is_parallel:
with concurrent.futures.ThreadPoolExecutor() as executor:
for spectral_density in executor.map(get_spectral_filter_worker, eigvec_ref_list, eigval_ref_list,
[filters for i in range(len(eigval_ref_list))],
[bound for i in range(len(eigval_ref_list))]):
sample_ref.append(spectral_density)
with concurrent.futures.ThreadPoolExecutor() as executor:
for spectral_density in executor.map(get_spectral_filter_worker, eigvec_pred_list, eigval_pred_list,
[filters for i in range(len(eigval_ref_list))],
[bound for i in range(len(eigval_ref_list))]):
sample_pred.append(spectral_density)
else:
for i in range(len(eigval_ref_list)):
try:
spectral_temp = get_spectral_filter_worker(eigvec_ref_list[i], eigval_ref_list[i], filters, bound)
sample_ref.append(spectral_temp)
except:
pass
for i in range(len(eigval_pred_list)):
try:
spectral_temp = get_spectral_filter_worker(eigvec_pred_list[i], eigval_pred_list[i], filters, bound)
sample_pred.append(spectral_temp)
except:
pass
if compute_emd:
# EMD option uses the same computation as GraphRNN, the alternative is MMD as computed by GRAN
# mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=emd)
mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=gaussian_emd)
else:
mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=gaussian_tv)
elapsed = datetime.now() - prev
if PRINT_TIME:
print('Time computing spectral filter stats: ', elapsed)
return mmd_dist
def spectral_stats(graph_ref_list, graph_pred_list, is_parallel=True, n_eigvals=-1, compute_emd=False):
''' Compute the distance between the degree distributions of two unordered sets of graphs.
Args:
graph_ref_list, graph_target_list: two lists of networkx graphs to be evaluated
'''
sample_ref = []
sample_pred = []
# in case an empty graph is generated
graph_pred_list_remove_empty = [
G for G in graph_pred_list if not G.number_of_nodes() == 0
]
prev = datetime.now()
if is_parallel:
with concurrent.futures.ThreadPoolExecutor() as executor:
for spectral_density in executor.map(spectral_worker, graph_ref_list, [n_eigvals for i in graph_ref_list]):
sample_ref.append(spectral_density)
with concurrent.futures.ThreadPoolExecutor() as executor:
for spectral_density in executor.map(spectral_worker, graph_pred_list_remove_empty,
[n_eigvals for i in graph_ref_list]):
sample_pred.append(spectral_density)
else:
for i in range(len(graph_ref_list)):
spectral_temp = spectral_worker(graph_ref_list[i], n_eigvals)
sample_ref.append(spectral_temp)
for i in range(len(graph_pred_list_remove_empty)):
spectral_temp = spectral_worker(graph_pred_list_remove_empty[i], n_eigvals)
sample_pred.append(spectral_temp)
# mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=gaussian_emd)
# mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=emd)
if compute_emd:
# EMD option uses the same computation as GraphRNN, the alternative is MMD as computed by GRAN
# mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=emd)
mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=gaussian_emd)
else:
mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=gaussian_tv)
# mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=gaussian)
elapsed = datetime.now() - prev
if PRINT_TIME:
print('Time computing degree mmd: ', elapsed)
return mmd_dist
###############################################################################
def clustering_worker(param):
G, bins = param
clustering_coeffs_list = list(nx.clustering(G).values())
hist, _ = np.histogram(
clustering_coeffs_list, bins=bins, range=(0.0, 1.0), density=False)
return hist
def clustering_stats(graph_ref_list,
graph_pred_list,
bins=100,
is_parallel=True, compute_emd=False):
sample_ref = []
sample_pred = []
graph_pred_list_remove_empty = [
G for G in graph_pred_list if not G.number_of_nodes() == 0
]
prev = datetime.now()
if is_parallel:
with concurrent.futures.ThreadPoolExecutor() as executor:
for clustering_hist in executor.map(clustering_worker,
[(G, bins) for G in graph_ref_list]):
sample_ref.append(clustering_hist)
with concurrent.futures.ThreadPoolExecutor() as executor:
for clustering_hist in executor.map(
clustering_worker, [(G, bins) for G in graph_pred_list_remove_empty]):
sample_pred.append(clustering_hist)
# check non-zero elements in hist
# total = 0
# for i in range(len(sample_pred)):
# nz = np.nonzero(sample_pred[i])[0].shape[0]
# total += nz
# print(total)
else:
for i in range(len(graph_ref_list)):
clustering_coeffs_list = list(nx.clustering(graph_ref_list[i]).values())
hist, _ = np.histogram(
clustering_coeffs_list, bins=bins, range=(0.0, 1.0), density=False)
sample_ref.append(hist)
for i in range(len(graph_pred_list_remove_empty)):
clustering_coeffs_list = list(
nx.clustering(graph_pred_list_remove_empty[i]).values())
hist, _ = np.histogram(
clustering_coeffs_list, bins=bins, range=(0.0, 1.0), density=False)
sample_pred.append(hist)
if compute_emd:
# EMD option uses the same computation as GraphRNN, the alternative is MMD as computed by GRAN
# mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=emd, sigma=1.0 / 10)
mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=gaussian_emd, sigma=1.0 / 10, distance_scaling=bins)
else:
mmd_dist = compute_mmd(sample_ref, sample_pred, kernel=gaussian_tv, sigma=1.0 / 10)
elapsed = datetime.now() - prev
if PRINT_TIME:
print('Time computing clustering mmd: ', elapsed)
return mmd_dist
# maps motif/orbit name string to its corresponding list of indices from orca output
motif_to_indices = {
'3path': [1, 2],
'4cycle': [8],
}
COUNT_START_STR = 'orbit counts:'
def edge_list_reindexed(G):
idx = 0
id2idx = dict()
for u in G.nodes():
id2idx[str(u)] = idx
idx += 1
edges = []
for (u, v) in G.edges():
edges.append((id2idx[str(u)], id2idx[str(v)]))
return edges
def orca(graph):
# tmp_fname = f'analysis/orca/tmp_{"".join(secrets.choice(ascii_uppercase + digits) for i in range(8))}.txt'
tmp_fname = f'orca/tmp_{"".join(secrets.choice(ascii_uppercase + digits) for i in range(8))}.txt'
tmp_fname = os.path.join(os.path.dirname(os.path.realpath(__file__)), tmp_fname)
# print(tmp_fname, flush=True)
f = open(tmp_fname, 'w')
f.write(
str(graph.number_of_nodes()) + ' ' + str(graph.number_of_edges()) + '\n')
for (u, v) in edge_list_reindexed(graph):
f.write(str(u) + ' ' + str(v) + '\n')
f.close()
output = sp.check_output(
[str(os.path.join(os.path.dirname(os.path.realpath(__file__)), 'orca/orca')), 'node', '4', tmp_fname, 'std'])
output = output.decode('utf8').strip()
idx = output.find(COUNT_START_STR) + len(COUNT_START_STR) + 2
output = output[idx:]
node_orbit_counts = np.array([
list(map(int,
node_cnts.strip().split(' ')))
for node_cnts in output.strip('\n').split('\n')
])
try:
os.remove(tmp_fname)
except OSError:
pass
return node_orbit_counts
def motif_stats(graph_ref_list, graph_pred_list, motif_type='4cycle', ground_truth_match=None,
bins=100, compute_emd=False):
# graph motif counts (int for each graph)
# normalized by graph size
total_counts_ref = []
total_counts_pred = []
num_matches_ref = []
num_matches_pred = []
graph_pred_list_remove_empty = [G for G in graph_pred_list if not G.number_of_nodes() == 0]
indices = motif_to_indices[motif_type]
for G in graph_ref_list:
orbit_counts = orca(G)
motif_counts = np.sum(orbit_counts[:, indices], axis=1)
if ground_truth_match is not None:
match_cnt = 0
for elem in motif_counts:
if elem == ground_truth_match:
match_cnt += 1
num_matches_ref.append(match_cnt / G.number_of_nodes())
# hist, _ = np.histogram(
# motif_counts, bins=bins, density=False)
motif_temp = np.sum(motif_counts) / G.number_of_nodes()
total_counts_ref.append(motif_temp)
for G in graph_pred_list_remove_empty:
orbit_counts = orca(G)
motif_counts = np.sum(orbit_counts[:, indices], axis=1)
if ground_truth_match is not None:
match_cnt = 0
for elem in motif_counts:
if elem == ground_truth_match:
match_cnt += 1
num_matches_pred.append(match_cnt / G.number_of_nodes())
motif_temp = np.sum(motif_counts) / G.number_of_nodes()
total_counts_pred.append(motif_temp)
total_counts_ref = np.array(total_counts_ref)[:, None]
total_counts_pred = np.array(total_counts_pred)[:, None]
if compute_emd:
# EMD option uses the same computation as GraphRNN, the alternative is MMD as computed by GRAN
# mmd_dist = compute_mmd(total_counts_ref, total_counts_pred, kernel=emd, is_hist=False)
mmd_dist = compute_mmd(total_counts_ref, total_counts_pred, kernel=gaussian, is_hist=False)
else:
mmd_dist = compute_mmd(total_counts_ref, total_counts_pred, kernel=gaussian, is_hist=False)
return mmd_dist
def orbit_stats_all(graph_ref_list, graph_pred_list, compute_emd=False):
total_counts_ref = []
total_counts_pred = []
graph_pred_list_remove_empty = [
G for G in graph_pred_list if not G.number_of_nodes() == 0
]
for G in graph_ref_list:
orbit_counts = orca(G)
orbit_counts_graph = np.sum(orbit_counts, axis=0) / G.number_of_nodes()
total_counts_ref.append(orbit_counts_graph)
for G in graph_pred_list:
orbit_counts = orca(G)
orbit_counts_graph = np.sum(orbit_counts, axis=0) / G.number_of_nodes()
total_counts_pred.append(orbit_counts_graph)
total_counts_ref = np.array(total_counts_ref)
total_counts_pred = np.array(total_counts_pred)
# mmd_dist = compute_mmd(
# total_counts_ref,
# total_counts_pred,
# kernel=gaussian,
# is_hist=False,
# sigma=30.0)
# mmd_dist = compute_mmd(
# total_counts_ref,
# total_counts_pred,
# kernel=gaussian_tv,
# is_hist=False,
# sigma=30.0)
if compute_emd:
# mmd_dist = compute_mmd(total_counts_ref, total_counts_pred, kernel=emd, sigma=30.0)
# EMD option uses the same computation as GraphRNN, the alternative is MMD as computed by GRAN
mmd_dist = compute_mmd(total_counts_ref, total_counts_pred, kernel=gaussian, is_hist=False, sigma=30.0)
else:
mmd_dist = compute_mmd(total_counts_ref, total_counts_pred, kernel=gaussian_tv, is_hist=False, sigma=30.0)
return mmd_dist
def eval_acc_lobster_graph(G_list):
G_list = [copy.deepcopy(gg) for gg in G_list]
count = 0
for gg in G_list:
if is_lobster_graph(gg):
count += 1
return count / float(len(G_list))
def eval_acc_tree_graph(G_list):
count = 0
for gg in G_list:
if nx.is_tree(gg):
count += 1
return count / float(len(G_list))
def eval_acc_grid_graph(G_list, grid_start=10, grid_end=20):
count = 0
for gg in G_list:
if is_grid_graph(gg):
count += 1
return count / float(len(G_list))
def eval_acc_sbm_graph(G_list, p_intra=0.3, p_inter=0.005, strict=True, refinement_steps=1000, is_parallel=True):
count = 0.0
if is_parallel:
with concurrent.futures.ThreadPoolExecutor() as executor:
for prob in executor.map(is_sbm_graph,
[gg for gg in G_list], [p_intra for i in range(len(G_list))],
[p_inter for i in range(len(G_list))],
[strict for i in range(len(G_list))],
[refinement_steps for i in range(len(G_list))]):
count += prob
else:
for gg in G_list:
count += is_sbm_graph(gg, p_intra=p_intra, p_inter=p_inter, strict=strict,
refinement_steps=refinement_steps)
return count / float(len(G_list))
def eval_acc_planar_graph(G_list):
count = 0
for gg in G_list:
if is_planar_graph(gg):
count += 1
return count / float(len(G_list))
def is_planar_graph(G):
return nx.is_connected(G) and nx.check_planarity(G)[0]
def is_lobster_graph(G):
"""
Check a given graph is a lobster graph or not
Removing leaf nodes twice:
lobster -> caterpillar -> path
"""
### Check if G is a tree
if nx.is_tree(G):
G = G.copy()
### Check if G is a path after removing leaves twice
leaves = [n for n, d in G.degree() if d == 1]
G.remove_nodes_from(leaves)
leaves = [n for n, d in G.degree() if d == 1]
G.remove_nodes_from(leaves)
num_nodes = len(G.nodes())
num_degree_one = [d for n, d in G.degree() if d == 1]
num_degree_two = [d for n, d in G.degree() if d == 2]
if sum(num_degree_one) == 2 and sum(num_degree_two) == 2 * (num_nodes - 2):
return True
elif sum(num_degree_one) == 0 and sum(num_degree_two) == 0:
return True
else:
return False
else:
return False
def is_grid_graph(G):
"""
Check if the graph is grid, by comparing with all the real grids with the same node count
"""
all_grid_file = f"data/all_grids.pt"
if os.path.isfile(all_grid_file):
all_grids = torch.load(all_grid_file)
else:
all_grids = {}
for i in range(2, 20):
for j in range(2, 20):
G_grid = nx.grid_2d_graph(i, j)
n_nodes = f"{len(G_grid.nodes())}"
all_grids[n_nodes] = all_grids.get(n_nodes, []) + [G_grid]
torch.save(all_grids, all_grid_file)
n_nodes = f"{len(G.nodes())}"
if n_nodes in all_grids:
for G_grid in all_grids[n_nodes]:
if nx.faster_could_be_isomorphic(G, G_grid):
if nx.is_isomorphic(G, G_grid):
return True
return False
else:
return False
# def is_sbm_graph(G, p_intra=0.3, p_inter=0.005, strict=True, refinement_steps=1000):
# """
# Check if how closely given graph matches a SBM with given probabilites by computing mean probability of Wald test statistic for each recovered parameter
# """
# adj = nx.adjacency_matrix(G).toarray()
# idx = adj.nonzero()
# g = gt.Graph()
# g.add_edge_list(np.transpose(idx))
# try:
# state = gt.minimize_blockmodel_dl(g)
# except ValueError:
# if strict:
# return False
# else:
# return 0.0
# # Refine using merge-split MCMC
# for i in range(refinement_steps):
# state.multiflip_mcmc_sweep(beta=np.inf, niter=10)
# b = state.get_blocks()
# b = gt.contiguous_map(state.get_blocks())
# state = state.copy(b=b)
# e = state.get_matrix()
# n_blocks = state.get_nonempty_B()
# node_counts = state.get_nr().get_array()[:n_blocks]
# edge_counts = e.todense()[:n_blocks, :n_blocks]
# if strict:
# if (node_counts > 40).sum() > 0 or (node_counts < 20).sum() > 0 or n_blocks > 5 or n_blocks < 2:
# return False
# max_intra_edges = node_counts * (node_counts - 1)
# est_p_intra = np.diagonal(edge_counts) / (max_intra_edges + 1e-6)
# max_inter_edges = node_counts.reshape((-1, 1)) @ node_counts.reshape((1, -1))
# np.fill_diagonal(edge_counts, 0)
# est_p_inter = edge_counts / (max_inter_edges + 1e-6)
# W_p_intra = (est_p_intra - p_intra) ** 2 / (est_p_intra * (1 - est_p_intra) + 1e-6)
# W_p_inter = (est_p_inter - p_inter) ** 2 / (est_p_inter * (1 - est_p_inter) + 1e-6)
# W = W_p_inter.copy()
# np.fill_diagonal(W, W_p_intra)
# p = 1 - chi2.cdf(abs(W), 1)
# p = p.mean()
# if strict:
# return p > 0.9 # p value < 10 %
# else:
# return p
def eval_fraction_isomorphic(fake_graphs, train_graphs):
count = 0
for fake_g in fake_graphs:
for train_g in train_graphs:
if nx.faster_could_be_isomorphic(fake_g, train_g):
if nx.is_isomorphic(fake_g, train_g):
count += 1
break
return count / float(len(fake_graphs))
def eval_fraction_unique(fake_graphs, precise=False):
count_non_unique = 0
fake_evaluated = []
for fake_g in fake_graphs:
unique = True
if not fake_g.number_of_nodes() == 0:
for fake_old in fake_evaluated:
if precise:
if nx.faster_could_be_isomorphic(fake_g, fake_old):
if nx.is_isomorphic(fake_g, fake_old):
count_non_unique += 1
unique = False
break
else:
if nx.faster_could_be_isomorphic(fake_g, fake_old):
if nx.could_be_isomorphic(fake_g, fake_old):
count_non_unique += 1
unique = False
break
if unique:
fake_evaluated.append(fake_g)
frac_unique = (float(len(fake_graphs)) - count_non_unique) / float(
len(fake_graphs)) # Fraction of distinct isomorphism classes in the fake graphs
return frac_unique
def eval_fraction_unique_non_isomorphic_valid(fake_graphs, train_graphs, validity_func=(lambda x: True)):
count_valid = 0
count_isomorphic = 0
count_non_unique = 0
fake_evaluated = []
for fake_g in fake_graphs:
unique = True
for fake_old in fake_evaluated:
if nx.faster_could_be_isomorphic(fake_g, fake_old):
if nx.is_isomorphic(fake_g, fake_old):
count_non_unique += 1
unique = False
break
if unique:
fake_evaluated.append(fake_g)
non_isomorphic = True
for train_g in train_graphs:
if nx.faster_could_be_isomorphic(fake_g, train_g):
if nx.is_isomorphic(fake_g, train_g):
count_isomorphic += 1
non_isomorphic = False
break
if non_isomorphic:
if validity_func(fake_g):
count_valid += 1
frac_unique = (float(len(fake_graphs)) - count_non_unique) / float(
len(fake_graphs)) # Fraction of distinct isomorphism classes in the fake graphs
frac_unique_non_isomorphic = (float(len(fake_graphs)) - count_non_unique - count_isomorphic) / float(
len(fake_graphs)) # Fraction of distinct isomorphism classes in the fake graphs that are not in the training set
frac_unique_non_isomorphic_valid = count_valid / float(
len(fake_graphs)) # Fraction of distinct isomorphism classes in the fake graphs that are not in the training set and are valid
return frac_unique, frac_unique_non_isomorphic, frac_unique_non_isomorphic_valid
class SpectreSamplingMetrics(nn.Module):
def __init__(self, data_loaders, compute_emd, metrics_list):
super().__init__()
self.train_graphs = self.loader_to_nx(data_loaders['train'])
self.val_graphs = self.loader_to_nx(data_loaders['val'])
self.test_graphs = self.loader_to_nx(data_loaders['test'])
self.num_graphs_test = len(self.test_graphs)
self.num_graphs_val = len(self.val_graphs)
self.compute_emd = compute_emd
self.metrics_list = metrics_list
def loader_to_nx(self, loader):
networkx_graphs = []
for i, batch in enumerate(loader):
data_list = batch.to_data_list()
for j, data in enumerate(data_list):
networkx_graphs.append(to_networkx(data, node_attrs=None, edge_attrs=None, to_undirected=True,
remove_self_loops=True))
return networkx_graphs
def forward(self, generated_graphs: list, local_rank, test=False):
reference_graphs = self.test_graphs if test else self.val_graphs
if local_rank == 0:
print(f"Computing sampling metrics between {len(generated_graphs)} generated graphs and {len(reference_graphs)}"
f" test graphs -- emd computation: {self.compute_emd}")
networkx_graphs = []
adjacency_matrices = []
if local_rank == 0:
print("Building networkx graphs...")
for graph in generated_graphs:
node_types, edge_types = graph
A = edge_types.bool().cpu().numpy()
adjacency_matrices.append(A)
nx_graph = nx.from_numpy_array(A)
networkx_graphs.append(nx_graph)
np.savez('generated_adjs.npz', *adjacency_matrices)
to_log = {}
if 'degree' in self.metrics_list:
if local_rank == 0:
print("Computing degree stats..")
degree = degree_stats(reference_graphs, networkx_graphs, is_parallel=True,
compute_emd=self.compute_emd)
to_log['degree'] = degree
if wandb.run:
wandb.run.summary['degree'] = degree
# val_eigvals = [graph["eigval"][1:self.k + 1].cpu().detach().numpy() for graph in self.val]
# train_eigvals = [graph["eigval"][1:self.k + 1].cpu().detach().numpy() for graph in self.train]
# eigval_stats(eig_ref_list, eig_pred_list, max_eig=20, is_parallel=True, compute_emd=False)
# spectral_filter_stats(eigvec_ref_list, eigval_ref_list, eigvec_pred_list, eigval_pred_list, is_parallel=False,
# compute_emd=False) # This is the one called wavelet
if 'spectre' in self.metrics_list:
if local_rank == 0:
print("Computing spectre stats...")
spectre = spectral_stats(reference_graphs, networkx_graphs, is_parallel=True, n_eigvals=-1,
compute_emd=self.compute_emd)
to_log['spectre'] = spectre
if wandb.run:
wandb.run.summary['spectre'] = spectre
if 'clustering' in self.metrics_list:
if local_rank == 0:
print("Computing clustering stats...")
clustering = clustering_stats(reference_graphs, networkx_graphs, bins=100, is_parallel=True,
compute_emd=self.compute_emd)
to_log['clustering'] = clustering
if wandb.run:
wandb.run.summary['clustering'] = clustering
if 'motif' in self.metrics_list:
if local_rank == 0:
print("Computing motif stats")
motif = motif_stats(reference_graphs, networkx_graphs, motif_type='4cycle', ground_truth_match=None, bins=100,
compute_emd=self.compute_emd)
to_log['motif'] = motif
if wandb.run:
wandb.run.summary['motif'] = motif
if 'orbit' in self.metrics_list:
if local_rank == 0:
print("Computing orbit stats...")
orbit = orbit_stats_all(reference_graphs, networkx_graphs, compute_emd=self.compute_emd)
to_log['orbit'] = orbit
if wandb.run:
wandb.run.summary['orbit'] = orbit
if 'sbm' in self.metrics_list:
if local_rank == 0:
print("Computing accuracy...")
acc = eval_acc_sbm_graph(networkx_graphs, refinement_steps=100, strict=True)
to_log['sbm_acc'] = acc
if wandb.run:
wandb.run.summary['sbmacc'] = acc
if 'planar' in self.metrics_list:
if local_rank ==0:
print('Computing planar accuracy...')
planar_acc = eval_acc_planar_graph(networkx_graphs)
to_log['planar_acc'] = planar_acc
if wandb.run:
wandb.run.summary['planar_acc'] = planar_acc
if 'sbm' or 'planar' in self.metrics_list:
if local_rank == 0:
print("Computing all fractions...")
frac_unique, frac_unique_non_isomorphic, fraction_unique_non_isomorphic_valid = eval_fraction_unique_non_isomorphic_valid(
networkx_graphs, self.train_graphs, is_sbm_graph if 'sbm' in self.metrics_list else is_planar_graph)
frac_non_isomorphic = 1.0 - eval_fraction_isomorphic(networkx_graphs, self.train_graphs)
to_log.update({'sampling/frac_unique': frac_unique,
'sampling/frac_unique_non_iso': frac_unique_non_isomorphic,
'sampling/frac_unic_non_iso_valid': fraction_unique_non_isomorphic_valid,
'sampling/frac_non_iso': frac_non_isomorphic})
if local_rank == 0:
print("Sampling statistics", to_log)
if wandb.run:
wandb.log(to_log, commit=False)
def reset(self):
pass
def loader_to_nx(loader):
networkx_graphs = {}
for i, batch in enumerate(loader):
data_list = batch.to_data_list()
for j, data in enumerate(data_list):
networkx_graphs[data.prompt_id.squeeze(0).item()] = [to_networkx(data, node_attrs=None, edge_attrs=None, to_undirected=True, remove_self_loops=True)]
return networkx_graphs
def compute_metrics(generated_graphs, referenced_graphs):
networkx_graphs = defaultdict(list)
adjacency_matrices = defaultdict(list)
for key in generated_graphs:
for graph in generated_graphs[key]:
node_types, edge_types = graph
A = edge_types.bool().cpu().numpy()
nx_graph = nx.from_numpy_array(A)
networkx_graphs[key].append(nx_graph)
adjacency_matrices[key].append(A)
new_referenced_graphs = []
for key in referenced_graphs:
new_referenced_graphs.extend(referenced_graphs[key])
referenced_graphs = new_referenced_graphs
nx_graphs = []
for key in networkx_graphs:
nx_graphs.extend(networkx_graphs[key])
return nx_graphs
class Comm20SamplingMetrics(SpectreSamplingMetrics):
def __init__(self, data_loaders):
super().__init__(data_loaders=data_loaders,
compute_emd=True,
metrics_list=['degree', 'clustering', 'orbit'])
class PlanarSamplingMetrics(SpectreSamplingMetrics):
def __init__(self, data_loaders):
super().__init__(data_loaders=data_loaders,
compute_emd=False,
metrics_list=['degree', 'clustering', 'orbit', 'spectre', 'planar'])
class SBMSamplingMetrics(SpectreSamplingMetrics):
def __init__(self, data_loaders):
super().__init__(data_loaders=data_loaders,
compute_emd=False,
metrics_list=['degree', 'clustering', 'orbit', 'spectre', 'sbm'])
class CrossDomainSamplingMetrics(SpectreSamplingMetrics):
def __init__(self, data_loaders):
super().__init__(data_loaders=data_loaders,
compute_emd=False,
metrics_list=['degree', 'clustering', 'orbit', 'spectre'])