experiment_num += 1
experiment_file_name = '{}-experiment-{}-results.json'.format(
network, experiment_num)
network_results_dir = './results/' + experiment_file_name
# Iterate over fractions of edges to hide
for frac_hidden in FRAC_EDGES_HIDDEN:
val_frac = 0.05
test_frac = frac_hidden - val_frac
# Read train-test split
experiment_name = '{}-{}-hidden'.format(network, frac_hidden)
print("Current experiment: ", experiment_name)
train_test_split_file = TRAIN_TEST_SPLITS_FOLDER + experiment_name + '.pkl'
# Run all link prediction methods on current graph, store results
network_results[experiment_name] = lp.calculate_all_scores(network_adj, features_matrix=None,
directed=False, \
test_frac=test_frac, val_frac=val_frac, \
random_state=RANDOM_SEED, verbose=2,
train_test_split_file=train_test_split_file,
tf_dtype=tf.float16)
# Save experiment results at each iteration
with open(network_results_dir, 'w') as fp:
json.dump(network_results, fp, indent=4)
# Save final experiment results
with open(network_results_dir, 'w') as fp:
json.dump(network_results, fp, indent=4)
feat = graph_tuple[1]
experiment_name = 'fb-{}-{}-hidden'.format(g_name, frac_hidden)
print "Current experiment: ", experiment_name
# # TODO: remove this!
# if experiment_name !='fb-combined-0.25-hidden' and \
# experiment_name != 'fb-combined-0.5-hidden' and \
# experiment_name != 'fb-combined-0.75-hidden':
# continue
train_test_split_file = TRAIN_TEST_SPLITS_FOLDER + experiment_name + '.pkl'
# Run all link prediction methods on current graph, store results
fb_results[experiment_name] = lp.calculate_all_scores(adj, feat, \
test_frac=test_frac, val_frac=val_frac, \
random_state=RANDOM_SEED, verbose=2,
train_test_split_file=train_test_split_file)
# Save experiment results at each iteration
with open(FB_RESULTS_DIR, 'wb') as f:
pickle.dump(fb_results, f, protocol=2)
# Save final experiment results
with open(FB_RESULTS_DIR, 'wb') as f:
pickle.dump(fb_results, f, protocol=2)
### ---------- NETWORKX ---------- ###
nx_results = {}
def facebook_networks():
### ---------- Load in FB Graphs ---------- ###
FB_EGO_USERS = [0, 107, 1684, 1912, 3437, 348, 3980, 414, 686, 698]
fb_graphs = {} # Dictionary to store all FB ego network graphs
# Read in each FB Ego graph
# Store graphs in dictionary as (adj, features) tuples
for user in FB_EGO_USERS:
network_dir = './data/fb-processed/{0}-adj-feat.pkl'.format(user)
with open(network_dir, 'rb') as f:
adj, features = pickle.load(f)
# Store in dict
fb_graphs[user] = (adj, features)
# Read in combined FB graph
combined_dir = './data/fb-processed/combined-adj-sparsefeat.pkl'
# with open(combined_dir, 'rb') as f:
# adj, features = pickle.load(f)
# fb_graphs['combined'] = (adj, features)
### ---------- Run Link Prediction Tests ---------- ###
for i in range(NUM_REPEATS):
## ---------- FACEBOOK ---------- ###
fb_results = {}
# Check existing experiment results, increment file number by 1
past_results = os.listdir('./results/')
txt_past_results = os.listdir('./results/txt/')
experiment_num = 0
experiment_file_name = 'fb-experiment-{}-results.pkl'.format(
experiment_num)
while (experiment_file_name in past_results):
experiment_num += 1
experiment_file_name = 'fb-experiment-{}-results.pkl'.format(
experiment_num)
FB_RESULTS_DIR = './results/' + experiment_file_name
# save as txt format
txt_experiment_num = 0
txt_experiment_file_name = 'txt-fb-experiment-{}-results.json'.format(
txt_experiment_num)
while (txt_experiment_file_name in txt_past_results):
txt_experiment_num += 1
txt_experiment_file_name = 'txt-fb-experiment-{}-results.json'.format(
txt_experiment_num)
TXT_FB_RESULTS_DIR = './results/txt/' + txt_experiment_file_name
TRAIN_TEST_SPLITS_FOLDER = './train-test-splits/'
# Iterate over fractions of edges to hide
for frac_hidden in FRAC_EDGES_HIDDEN:
val_frac = 0.05
test_frac = frac_hidden - val_frac
# Iterate over each graph
for g_name, graph_tuple in fb_graphs.items():
adj = graph_tuple[0]
feat = graph_tuple[1]
experiment_name = 'fb-{}-{}-hidden'.format(g_name, frac_hidden)
print("Current experiment: ", experiment_name)
# # TODO: remove this!
# if experiment_name !='fb-combined-0.25-hidden' and \
# experiment_name != 'fb-combined-0.5-hidden' and \
# experiment_name != 'fb-combined-0.75-hidden':
# continue
train_test_split_file = TRAIN_TEST_SPLITS_FOLDER + experiment_name + '.pkl'
# Run all link prediction methods on current graph, store results
fb_results[experiment_name] = lp.calculate_all_scores(adj, feat, \
test_frac=test_frac, val_frac=val_frac, \
random_state=RANDOM_SEED, verbose=2,
train_test_split_file=train_test_split_file)
# Save experiment results at each iteration
with open(FB_RESULTS_DIR, 'wb') as f:
pickle.dump(fb_results, f, protocol=2)
# with open(TXT_FB_RESULTS_DIR, 'w') as f:
# json.dump(fb_results, f, indent=4)
# Save final experiment results
with open(FB_RESULTS_DIR, 'wb') as f:
pickle.dump(fb_results, f, protocol=2)
def random_networks():
### ---------- Create Random NetworkX Graphs ---------- ###
# Dictionary to store all nx graphs
nx_graphs = {}
# Small graphs
# N_SMALL = 200
# nx_graphs['er-small'] = nx.erdos_renyi_graph(n=N_SMALL, p=.03, seed=RANDOM_SEED) # Erdos-Renyi
# nx_graphs['ws-small'] = nx.watts_strogatz_graph(n=N_SMALL, k=11, p=.1, seed=RANDOM_SEED) # Watts-Strogatz
# nx_graphs['ba-small'] = nx.barabasi_albert_graph(n=N_SMALL, m=6, seed=RANDOM_SEED) # Barabasi-Albert
# nx_graphs['pc-small'] = nx.powerlaw_cluster_graph(n=N_SMALL, m=6, p=.02, seed=RANDOM_SEED) # Powerlaw Cluster
# nx_graphs['sbm-small'] = nx.random_partition_graph(sizes=[N_SMALL//10]*10, p_in=.1, p_out=.01, seed=RANDOM_SEED) # Stochastic Block Model
# Larger graphs
NUM = [12, 100, 1000, 10000]
for N_LARGE in NUM:
# N_LARGE = 1000
# nx_graphs['er-large'] = nx.erdos_renyi_graph(n=N_LARGE, p=.03, seed=RANDOM_SEED) # Erdos-Renyi
# nx_graphs['ws-large'] = nx.watts_strogatz_graph(n=N_LARGE, k=11, p=.1, seed=RANDOM_SEED) # Watts-Strogatz
nx_graphs['ba-large'] = nx.barabasi_albert_graph(
n=N_LARGE, m=10, seed=RANDOM_SEED) # Barabasi-Albert
# nx_graphs['pc-large'] = nx.powerlaw_cluster_graph(n=N_LARGE, m=6, p=.02, seed=RANDOM_SEED) # Powerlaw Cluster
# nx_graphs['sbm-large'] = nx.random_partition_graph(sizes=[N_LARGE//10]*10, p_in=.05, p_out=.005, seed=RANDOM_SEED) # Stochastic Block Model
# Remove isolates from random graphs
for g_name, nx_g in nx_graphs.items():
isolates = nx.isolates(nx_g)
if len(list(isolates)) > 0:
for isolate_node in isolates:
nx_graphs[g_name].remove_node(isolate_node)
### ---------- Run Link Prediction Tests ---------- ###
for i in range(NUM_REPEATS):
## ---------- NETWORKX ---------- ###
nx_results = {}
# Check existing experiment results, increment file number by 1
past_results = os.listdir('./results/')
txt_past_results = os.listdir('./results/txt/')
experiment_num = 0
experiment_file_name = 'nx-experiment-{}-results.pkl'.format(
experiment_num)
while (experiment_file_name in past_results):
experiment_num += 1
experiment_file_name = 'nx-experiment-{}-results.pkl'.format(
experiment_num)
NX_RESULTS_DIR = './results/' + experiment_file_name
# save as txt format
txt_experiment_num = 0
txt_experiment_file_name = 'txt-nx-experiment-{}-results.json'.format(
txt_experiment_num)
while (txt_experiment_file_name in txt_past_results):
txt_experiment_num += 1
txt_experiment_file_name = 'txt-nx-experiment-{}-results.json'.format(
txt_experiment_num)
TXT_NX_RESULTS_DIR = './results/txt/' + txt_experiment_file_name
# Iterate over fractions of edges to hide
for frac_hidden in FRAC_EDGES_HIDDEN:
val_frac = 0.05
test_frac = frac_hidden - val_frac
# Iterate over each graph
for g_name, nx_g in nx_graphs.items():
adj = nx.adjacency_matrix(nx_g)
experiment_name = 'nx-{}-{}-hidden'.format(
g_name, frac_hidden)
print("Current experiment: ", experiment_name)
# Run all link prediction methods on current graph, store results
nx_results[experiment_name] = lp.calculate_all_scores(adj, \
test_frac=test_frac, val_frac=val_frac, \
random_state=RANDOM_SEED, verbose=0)
# Save experiment results at each iteration
with open(NX_RESULTS_DIR, 'wb') as f:
pickle.dump(nx_results, f, protocol=2)
with open(TXT_NX_RESULTS_DIR, 'w') as f:
json.dump(nx_results, f, indent=4)
# Save final experiment results
with open(NX_RESULTS_DIR, 'wb') as f:
pickle.dump(nx_results, f, protocol=2)
with open(TXT_NX_RESULTS_DIR, 'w+') as f:
json.dump(nx_results, f, indent=4)
def facebook_networks():
###---------- 读取Facebook网络数据----------###
FB_EGO_USERS = [0, 107, 1684, 1912, 3437, 348, 3980, 414, 686, 698]
fb_graphs = {} # 保存所有的FB自我中心网络
# 读取每个FB自我中心网络
for user in FB_EGO_USERS:
network_dir = './data/fb-processed/{0}-adj-feat.pkl'.format(user)
with open(network_dir, 'rb') as f:
adj, features = pickle.load(f)
# 保存在字典目录
fb_graphs[user] = (adj, features)
# 读取FB-combined网络
combined_dir = './data/fb-processed/combined-adj-sparsefeat.pkl'
with open(combined_dir, 'rb') as f:
adj, features = pickle.load(f)
fb_graphs['combined'] = (adj, features)
### ---------- 运行链路预测实验 ---------- ###
for i in range(NUM_REPEATS):
fb_results = {}
# 根据实验命名实验结果文件
past_results = os.listdir('./results/')
txt_past_results = os.listdir('./results/txt/')
experiment_num = 0
experiment_file_name = 'fb-experiment-{}-results.pkl'.format(experiment_num)
while (experiment_file_name in past_results):
experiment_num += 1
experiment_file_name = 'fb-experiment-{}-results.pkl'.format(experiment_num)
FB_RESULTS_DIR = './results/' + experiment_file_name
# 保存为txt文件
txt_experiment_num = 0
txt_experiment_file_name = 'txt-fb-experiment-{}-results.json'.format(txt_experiment_num)
while (txt_experiment_file_name in txt_past_results):
txt_experiment_num += 1
txt_experiment_file_name = 'txt-fb-experiment-{}-results.json'.format(txt_experiment_num)
TXT_FB_RESULTS_DIR = './results/txt/' + txt_experiment_file_name
TRAIN_TEST_SPLITS_FOLDER = './train-test-splits/'
# 遍历不同隐藏比例的数据集（训练集、验证集、测试集）
for frac_hidden in FRAC_EDGES_HIDDEN:
val_frac = 0.05
test_frac = frac_hidden - val_frac
# 遍历每个网络集
for g_name, graph_tuple in fb_graphs.items():
adj = graph_tuple[0]
feat = graph_tuple[1]
experiment_name = 'fb-{}-{}-hidden'.format(g_name, frac_hidden)
print("Current experiment: ", experiment_name)
train_test_split_file = TRAIN_TEST_SPLITS_FOLDER + experiment_name + '.pkl'
# 在当前网络上运行所有链接预测方法，返回结果
fb_results[experiment_name] = lp.calculate_all_scores(adj, feat, \
test_frac=test_frac, val_frac=val_frac, \
random_state=RANDOM_SEED, verbose=2,
train_test_split_file=train_test_split_file)
# 每次遍历保存实验结果
# pickle文件
with open(FB_RESULTS_DIR, 'wb') as f:
pickle.dump(fb_results, f, protocol=2)
# json文件
with open(TXT_FB_RESULTS_DIR, 'w') as f:
json.dump(fb_results, f, indent=4)
# 保存最终实验结果
# pickle文件
with open(FB_RESULTS_DIR, 'wb') as f:
pickle.dump(fb_results, f, protocol=2)
# json文件
with open(TXT_FB_RESULTS_DIR, 'w') as f:
json.dump(fb_results, f, indent=4)
def random_networks():
### ---------- 生成随机网络图 ---------- ###
# 保存生成的随机网络
nx_graphs = {}
#NUM=[10, 100, 1000, 10000]
NUM = [2000]
for N_LARGE in NUM:
# N_LARGE = 1000
# nx_graphs['er-large'] = nx.erdos_renyi_graph(n=N_LARGE, p=.03, seed=RANDOM_SEED) # Erdos-Renyi
# nx_graphs['ws-large'] = nx.watts_strogatz_graph(n=N_LARGE, k=11, p=.1, seed=RANDOM_SEED) # Watts-Strogatz
nx_graphs['ba-large'] = nx.barabasi_albert_graph(n=N_LARGE, m=6, seed=RANDOM_SEED) # Barabasi-Albert
# nx_graphs['pc-large'] = nx.powerlaw_cluster_graph(n=N_LARGE, m=6, p=.02, seed=RANDOM_SEED) # Powerlaw Cluster
# nx_graphs['sbm-large'] = nx.random_partition_graph(sizes=[N_LARGE//10]*10, p_in=.05, p_out=.005, seed=RANDOM_SEED) # Stochastic Block Model
# 移除孤立点
for g_name, nx_g in nx_graphs.items():
isolates = nx.isolates(nx_g)
if len(list(isolates)) > 0:
for isolate_node in isolates:
nx_graphs[g_name].remove_node(isolate_node)
### ---------- 运行链路预测实验 ---------- ###
for i in range(NUM_REPEATS):
## ---------- NETWORKX ---------- ###
nx_results = {}
# 根据实验命名实验结果文件
past_results = os.listdir('./results/')
txt_past_results = os.listdir('./results/txt/')
experiment_num = 0
experiment_file_name = 'nx-experiment-{}-results.pkl'.format(experiment_num)
while (experiment_file_name in past_results):
experiment_num += 1
experiment_file_name = 'nx-experiment-{}-results.pkl'.format(experiment_num)
NX_RESULTS_DIR = './results/' + experiment_file_name
# 保存为txt文件
txt_experiment_num = 0
txt_experiment_file_name = 'txt-nx-experiment-{}-results.json'.format(txt_experiment_num)
while (txt_experiment_file_name in txt_past_results):
txt_experiment_num += 1
txt_experiment_file_name = 'txt-nx-experiment-{}-results.json'.format(txt_experiment_num)
TXT_NX_RESULTS_DIR = './results/txt/' + txt_experiment_file_name
# 遍历不同隐藏比例的数据集（训练集、验证集、测试集）
for frac_hidden in FRAC_EDGES_HIDDEN:
val_frac = 0.05
test_frac = frac_hidden - val_frac
# 遍历每个随机网络
for g_name, nx_g in nx_graphs.items():
adj = nx.adjacency_matrix(nx_g)
experiment_name = 'nx-{}-{}-hidden'.format(g_name, frac_hidden)
print("Current experiment: ", experiment_name)
# 在当前网络上运行所有链接预测方法，返回结果
nx_results[experiment_name] = lp.calculate_all_scores(adj, \
test_frac=test_frac, val_frac=val_frac, \
random_state=RANDOM_SEED, verbose=0)
# 每次遍历保存实验结果
with open(NX_RESULTS_DIR, 'wb') as f:
pickle.dump(nx_results, f, protocol=2)
with open(TXT_NX_RESULTS_DIR, 'w') as f:
json.dump(nx_results, f, indent=4)
# 保存最终实验结果
with open(NX_RESULTS_DIR, 'wb') as f:
pickle.dump(nx_results, f, protocol=2)
with open(TXT_NX_RESULTS_DIR, 'w+') as f:
json.dump(nx_results, f, indent=4)
while (experiment_file_name in past_results):
experiment_num += 1
experiment_file_name = 'twitter-experiment-{}-results.json'.format(experiment_num)
twitter_results_dir = './results/' + experiment_file_name
# Iterate over fractions of edges to hide
for frac_hidden in FRAC_EDGES_HIDDEN:
val_frac = 0.1
test_frac = frac_hidden - val_frac
# Read train-test split
experiment_name = 'twitter-combined-{}-hidden'.format(frac_hidden)
print "Current experiment: ", experiment_name
train_test_split_file = TRAIN_TEST_SPLITS_FOLDER + experiment_name + '.pkl'
# Run all link prediction methods on current graph, store results
twitter_results[experiment_name] = lp.calculate_all_scores(twitter_adj, features_matrix=None,
directed=True, \
test_frac=test_frac, val_frac=val_frac, \
random_state=RANDOM_SEED, verbose=2,
train_test_split_file=train_test_split_file,
tf_dtype=tf.float16)
# Save experiment results at each iteration
with open(twitter_results_dir, 'w') as fp:
json.dump(twitter_results, fp, indent=4)
# Save final experiment results
with open(twitter_results_dir, 'w') as fp:
json.dump(twitter_results, fp, indent=4)