def compute_graph_stats(root):
print('computing GraphStats... ', end='', flush=True)
if type(root) is list:
graph_stats = [GraphStats(graph=g, run_id = 1) for g in root]
else:
graph_stats = [GraphStats(graph=node.graph, run_id=1) for node in [root] + list(root.descendants)]
print('done')
return graph_stats
def main():
base_path = '/data/infinity-mirror/'
input_path = '/home/dgonza26/infinity-mirror/input'
dataset = 'flights'
model = 'GraphRNN'
output_path = os.path.join(base_path, 'stats', 'degree')
abs_lambda = []
trials = []
R = [(root, generation) for root, generation in load_data(
os.path.join(base_path, 'cleaned'), dataset)]
R.sort(key=lambda x: x[1])
R = [root for (root, generation) in R]
if dataset == 'clique-ring-500-4':
G = nx.ring_of_cliques(500, 4)
else:
G = init(os.path.join(input_path, f'{dataset}.g'))
# add the initial graph and transpose the list
roots = [list(r) for r in zip(*R)]
cols = ['model', 'gen', 'abs']
rows = {col: [] for col in cols}
gs0 = GraphStats(graph=G, run_id=1)
for i, chain in enumerate(roots, 1):
print(f'chain: {i}')
for idx, graph in enumerate(chain, 1):
print(f'\tgen: {idx} ... ', end='', flush=True)
comparator = GraphPairCompare(gs0, GraphStats(graph=graph,
run_id=1))
try:
rows['abs'].append(comparator.js_distance())
except Exception as e:
print(f'ERROR\n{e}')
else:
rows['model'].append('GraphRNN')
rows['gen'].append(idx)
print('done')
df = pd.DataFrame(rows)
print(df.head())
df.to_csv(f'{output_path}/{dataset}_{model}_lambda.csv',
float_format='%.7f',
sep='\t',
index=False,
na_rep='nan')
print(f'wrote {output_path}/{dataset}_{model}_lambda.csv')
#df.to_csv(f'{output_path}/{dataset}_{model}_lambda.csv', float_format='%.7f', sep='\t', index=False, na_rep='nan')
return
def _fit(self) -> None:
# find degree distribution and avg clustering by degree
g_stats = GraphStats(self.input_graph, trial=-1)
self.params['n'] = self.input_graph.order()
self.params['degree_dist'] = g_stats.degree_dist(
normalized=False) # we need the counts
self.params['degree_seq'] = g_stats['degree_seq']
self.params[
'avg_cc_by_deg'] = g_stats.clustering_coefficients_by_degree()
return
def stats_computation(dataset, model, trial, filename, stats):
path = Path(
get_imt_output_directory()) / 'pickles' / dataset / model / filename
graph_list = load_pickle(path)
assert isinstance(
graph_list,
list), f'Expected type "list" and got type {type(graph_list)}.'
assert all(isinstance(g, nx.Graph) for g in graph_list
), f'Expected a list of nx.Graph and got disappointed instead.'
ColorPrint.print_orange(f'{filename} has length {len(graph_list)}')
for idx, G in enumerate(graph_list):
gs_obj = GraphStats(graph=G,
dataset=dataset,
model=model,
trial=trial,
iteration=idx)
gs_obj.write_stats_jsons(stats=stats)
return None
if __name__ == '__main__':
base_path = '/data/infinity-mirror/'
dataset = 'clique-ring-500-4'
models = ['CNRG', 'Chung-Lu', 'Erdos-Renyi', 'SBM', 'HRG', 'NetGAN']
output_path = os.path.join(base_path, 'stats/deltacon/')
mkdir_output(output_path)
cols = ['model', 'gen', 'trial_id', 'deltacon']
g1 = nx.ring_of_cliques(500, 4)
g2 = nx.ring_of_cliques(500, 4)
gpc1 = GraphPairCompare(GraphStats(g1, 1), GraphStats(g1, 1))
print('should def be 0', gpc1.deltacon0())
gpc2 = GraphPairCompare(GraphStats(g1, 1), GraphStats(g2, 1))
print('should probably be 0', gpc2.deltacon0())
exit()
for model in models[:1]:
abs_delta = []
trials = []
rows = {col: [] for col in cols}
for root, trial in load_data(base_path, dataset, model, True, False):
# graph_stats = compute_graph_stats(root)
if isinstance(root, list):
root_graph = root[0]
def compute_graph_stats(root):
graph_stats = [
GraphStats(graph=node.graph, run_id=1)
for node in [root] + list(root.descendants)
]
return graph_stats
for filename in files:
string = subdir.split('/')[-2:]
file = os.path.join(subdir, filename)
if 'seq' in file and 'rob' not in file:
newfile = file.split('.')[0] + '_rob.pkl.gz'
#newfile = file[:-7] + '_rob.pkl.gz'
print(f'starting\t{string[-2]}\t{string[-1]}\t{filename} ... ')
root = load_pickle(file)
try:
root.robustness
except AttributeError:
root = TreeNode(name=root.name, graph=root.graph, stats=root.stats, stats_seq=root.stats_seq, parent=root.parent, children=root.children)
if root.robustness is None or root.robustness == {}:
graphstats = GraphStats(graph=root.graph, run_id=0)
graphstats._calculate_robustness_measures()
root.robustness = graphstats.stats
for node in root.descendants:
try:
node.robustness
except AttributeError:
node = TreeNode(name=node.name, graph=node.graph, stats=node.stats, stats_seq=node.stats_seq, parent=node.parent, children=node.children)
if node.robustness is None or node.robustness == {}:
graphstats = GraphStats(graph=node.graph, run_id=0)
graphstats._calculate_robustness_measures()
node.robustness = graphstats.stats
with open(newfile, 'wb') as f:
pickle.dump(root, f)
print('done')
def main(base_path, dataset, models):
if 'GraphRNN' in models:
#path = os.path.join(base_path, 'GraphRNN')
#for subdir, dirs, files in os.walk(path):
# if dataset == subdir.split('/')[-1].split('_')[0]:
# print(subdir)
# for filename in files:
# print(filename)
models.remove('GraphRNN')
for model in models:
path = os.path.join(base_path, dataset, model)
for subdir, dirs, files in os.walk(path):
for filename in files:
if 'seq' not in filename:
run_id = int(filename.split('.')[0].split('_')[-1])
string = subdir.split('/')[-2:]
file = os.path.join(subdir, filename)
newfile = file.split('.')[0]
if 'rob' in file:
newfile += '_seq_rob.pkl.gz'
else:
newfile += '_seq.pkl.gz'
print(f'starting\t{string[-2]}\t{string[-1]}\t{filename} ... ', end='', flush=True)
root = load_pickle(file)
node = root
try:
node.stats_seq
except AttributeError:
if type(node) is LightTreeNode:
node_graph_stats = GraphStats(run_id=run_id, graph=node.graph)
comparator = GraphPairCompare(GraphStats(graph=root.graph, run_id=run_id), \
GraphStats(graph=root.graph, run_id=run_id))
stats = {}
stats['lambda_dist'] = comparator.lambda_dist()
stats['node_diff'] = comparator.node_diff()
stats['edge_diff'] = comparator.edge_diff()
stats['pgd_pearson'] = comparator.pgd_pearson()
stats['pgd_spearman'] = comparator.pgd_spearman()
stats['deltacon0'] = comparator.deltacon0()
stats['degree_cvm'] = comparator.cvm_degree()
stats['pagerank_cvm'] = comparator.cvm_pagerank()
node = TreeNode(name=node.name, graph=node.graph, stats=stats, stats_seq={}, parent=node.parent, children=node.children)
elif type(node) is TreeNode:
node = TreeNode(name=node.name, graph=node.graph, stats=node.stats, stats_seq={}, parent=node.parent, children=node.children)
else:
print(f'node has unknown type: {type(node)}')
exit()
if node.stats_seq is None or node.stats_seq == {}:
node.stats_seq = {}
while len(node.children) > 0:
child = node.children[0]
try:
child.stats_seq = {}
except AttributeError:
child = TreeNode(name=child.name, graph=child.graph, stats=child.stats, stats_seq={}, parent=child.parent, children=child.children)
comparator = GraphPairCompare(GraphStats(graph=node.graph, run_id=run_id), \
GraphStats(graph=child.graph, run_id=run_id))
child.stats_seq['lambda_dist'] = comparator.lambda_dist()
child.stats_seq['node_diff'] = comparator.node_diff()
child.stats_seq['edge_diff'] = comparator.edge_diff()
child.stats_seq['pgd_pearson'] = comparator.pgd_pearson()
child.stats_seq['pgd_spearman'] = comparator.pgd_spearman()
child.stats_seq['deltacon0'] = comparator.deltacon0()
child.stats_seq['degree_cvm'] = comparator.cvm_degree()
child.stats_seq['pagerank_cvm'] = comparator.cvm_pagerank()
node = child
with open(newfile, 'wb') as f:
pickle.dump(root, f)
print(f'\tdone')
def main():
base_path = '/data/infinity-mirror/'
input_path = '/home/dgonza26/infinity-mirror/input'
dataset = 'flights'
model = 'GraphRNN'
output_path = os.path.join(base_path, 'stats', 'pgd')
abs_lambda = []
trials = []
R = [(root, generation) for root, generation in load_data(
os.path.join(base_path, 'cleaned'), dataset)]
R.sort(key=lambda x: x[1])
R = [root for (root, generation) in R]
if dataset == 'clique-ring-500-4':
G = nx.ring_of_cliques(500, 4)
else:
G = init(os.path.join(input_path, f'{dataset}.g'))
# add the initial graph and transpose the list
roots = [[G] + list(r) for r in zip(*R)]
cols = [
'model', 'gen', 'trial', 'total_2_1edge', 'total_2_indep',
'total_3_tris', 'total_2_star', 'total_3_1edge', 'total_4_clique',
'total_4_chordcycle', 'total_4_tailed_tris', 'total_3_star',
'total_4_path', 'total_4_1edge', 'total_4_2edge', 'total_4_2star',
'total_4_tri', 'total_4_indep'
]
rows = {col: [] for col in cols}
for i, chain in enumerate(roots, 1):
print(f'chain: {i}')
for idx, graph in enumerate(chain):
print(f'\tgen: {idx} ... ', end='', flush=True)
try:
pgd = GraphStats(graph=graph, run_id=1).pgd_graphlet_counts()
except Exception as e:
print(f'ERROR\n{e}')
else:
if pgd == {} and len(graph.edges()) == 0:
rows['model'].append('GraphRNN')
rows['gen'].append(idx)
rows['trial'].append(i)
rows['total_2_1edge'].append(0)
rows['total_2_indep'].append(0)
rows['total_3_tris'].append(0)
rows['total_2_star'].append(0)
rows['total_3_1edge'].append(0)
rows['total_4_clique'].append(0)
rows['total_4_chordcycle'].append(0)
rows['total_4_tailed_tris'].append(0)
rows['total_3_star'].append(0)
rows['total_4_path'].append(0)
rows['total_4_1edge'].append(0)
rows['total_4_2edge'].append(0)
rows['total_4_2star'].append(0)
rows['total_4_tri'].append(0)
rows['total_4_indep'].append(0)
print('done')
else:
rows['model'].append('GraphRNN')
rows['gen'].append(idx)
rows['trial'].append(i)
rows['total_2_1edge'].append(pgd['total_2_1edge'])
rows['total_2_indep'].append(pgd['total_2_indep'])
rows['total_3_tris'].append(pgd['total_3_tris'])
rows['total_2_star'].append(pgd['total_2_star'])
rows['total_3_1edge'].append(pgd['total_3_1edge'])
rows['total_4_clique'].append(pgd['total_4_clique'])
rows['total_4_chordcycle'].append(
pgd['total_4_chordcycle'])
rows['total_4_tailed_tris'].append(
pgd['total_4_tailed_tris'])
rows['total_3_star'].append(pgd['total_3_star'])
rows['total_4_path'].append(pgd['total_4_path'])
rows['total_4_1edge'].append(pgd['total_4_1edge'])
rows['total_4_2edge'].append(pgd['total_4_2edge'])
rows['total_4_2star'].append(pgd['total_4_2star'])
rows['total_4_tri'].append(pgd['total_4_tri'])
rows['total_4_indep'].append(pgd['total_4_indep'])
print('done')
df = pd.DataFrame(rows)
print(df.head())
df.to_csv(f'{output_path}/{dataset}_{model}_pgd_full.csv',
float_format='%.7f',
sep='\t',
index=False,
na_rep='nan')
print(f'wrote {output_path}/{dataset}_{model}_pgd_full.csv')
#df.to_csv(f'{output_path}/{dataset}_{model}_lambda.csv', float_format='%.7f', sep='\t', index=False, na_rep='nan')
return