def para_replace(city, defense_name, checkin, ratio, step):
''' para replace
Args:
city: city
defense_name: 20_replace_20
checkin: checkin data
ratio: obfuscate proportion, 0.1, 0.2, 0.3 ...
step: step length, (odd)
Returns:
checkin: checkin data
'''
# intialize
pd.DataFrame([checkin.columns]).to_csv('dataset/'+city+'/defense/'+\
city+'_'+defense_name+'.checkin', index=False, header=None)
ul_graph, lu_graph = ul_graph_build(checkin, 'locid')
core_num = mp.cpu_count()
print(core_num)
Parallel(n_jobs = core_num)(delayed(replace_core)(\
city, defense_name, checkin.loc[checkin.uid==u], ul_graph, lu_graph, ratio, step) for u in checkin.uid.unique())
checkin = pd.read_csv('dataset/'+city+'/defense/'+city+'_'+defense_name+'.checkin',\
error_bad_lines=False)
checkin = checkin.dropna().reset_index(drop=True)
checkin.uid = checkin.uid.apply(float).apply(int)
checkin.locid = checkin.locid.apply(float).apply(int)
checkin = checkin.reset_index(drop=True)
checkin.to_csv('dataset/'+city+'/defense/'+city+'_'+defense_name+'.checkin',\
index=False)
return checkin
def single_run(city, cicnt=20, wl=100, wt=20, n_feature=128, new_run=False):
# city = "Brightkite"# ny la london Gowalla Brightkite
# cicnt = 20
folder_setup(city)
checkin, friends = data_process(city, cicnt)
ul_graph, lu_graph = ul_graph_build(checkin, 'locid')
model_name = str(cicnt) + '_locid'
print(model_name)
walk_len, walk_times = 100, 20 # maximal 100 walk_len, 20 walk_times
print('walking')
if new_run:
para_ul_random_batch(city, model_name, checkin.uid.unique(), ul_graph,
lu_graph, walk_len, walk_times)
print('walk done')
print('emb training')
emb_train(city, model_name, wl, wt, n_feature)
print('emb training done')
feature_construct(city, model_name, friends, wl, wt, n_feature)
unsuper_friends_predict(city, model_name, wl, wt, n_feature)
def single_run(city, cicnt, ratio):
# ratio = int(sys.argv[3])# 10 20 30 40
ratio = ratio * 1.0 / 100
folder_setup(city)
checkin, friends = data_process(city, cicnt)
defense_name = str(cicnt) + '_hiding_' + str(int(ratio * 100))
print(defense_name)
checkin = para_hiding(city, defense_name, checkin, ratio)
ul_graph, lu_graph = ul_graph_build(checkin, 'locid')
model_name = str(cicnt) + '_locid_hiding_' + str(int(ratio * 100))
print(model_name)
walk_len, walk_times = 100, 20 # maximal 100 walk_len, 20 walk_times
print('walking')
para_ul_random_walk(city, model_name, checkin.uid.unique(), ul_graph,
lu_graph, walk_len, walk_times)
print('walk done')
print('emb training')
emb_train(city, model_name)
print('emb training done')
feature_construct(city, model_name, friends)
unsuper_friends_predict(city, model_name)
def single_ex_run(city, cicnt):
folder_setup(city)
checkin, friends = data_process(city, cicnt)
defense_name = str(cicnt) + '_ex'
print(defense_name)
checkin = extreme_balance(city, defense_name, checkin)
ul_graph, lu_graph = ul_graph_build(checkin, 'locid')
model_name = str(cicnt) + '_locid_ex'
print(model_name)
walk_len, walk_times = 100, 20 # maximal 100 walk_len, 20 walk_times
print('walking')
para_ul_random_batch(city, model_name, checkin.uid.unique(), ul_graph,
lu_graph, walk_len, walk_times)
print('walk done')
print('emb training')
emb_train(city, model_name)
print('emb training done')
feature_construct(city, model_name, friends)
unsuper_friends_predict(city, model_name)
def batch_random_walk(num_user, num_location):
city = 'workload_' + str(num_user) + '_' + str(num_location)
folder_setup(city)
cicnt = 20
checkin, friends = data_process(city, cicnt)
ul_graph, lu_graph = ul_graph_build(checkin, 'locid')
model_name = str(cicnt) + '_locid'
print(model_name)
walk_len, walk_times = 50, 10 # maximal 100 walk_len, 20 walk_times
print('walking')
para_ul_random_batch(city, model_name, checkin.uid.unique(),
ul_graph, lu_graph, walk_len, walk_times)
print('walk done')
def single_replace(city, cicnt, ratio, step, fail_to_continue=False):
ratio = ratio * 1.0 / 100
folder_setup(city)
checkin, friends = data_process(city, cicnt)
defense_name = str(cicnt) + '_replace_' + str(int(
ratio * 100)) + '_' + str(int(step))
model_name = str(cicnt) + '_locid_replace_' + str(int(
ratio * 100)) + '_' + str(int(step))
if not fail_to_continue:
checkin = para_replace(city, defense_name, checkin, ratio, step)
else:
checkin = pd.read_csv('dataset/'+ city + '/defense/' + city + \
'_20_replace_'+
str(int(ratio * 100)) + '_' + str(int(step)) +
'.checkin')
ul_graph, lu_graph = ul_graph_build(checkin, 'locid')
walk_len, walk_times = 100, 20 # maximal 100 walk_len, 20 walk_times
print('walking')
if not fail_to_continue:
para_ul_random_batch(city, model_name, checkin.uid.unique(), ul_graph,
lu_graph, walk_len, walk_times)
print('walk done')
print('emb training')
emb_train(city, model_name)
print('emb training done')
feature_construct(city, model_name, friends)
unsuper_friends_predict(city, model_name)
city = sys.argv[1]
cicnt = int(sys.argv[2])
ratio = int(sys.argv[3]) # 10 20 30 40
ratio = ratio * 1.0 / 100
step = int(sys.argv[4])
folder_setup(city)
checkin, friends = data_process(city, cicnt)
defense_name = str(cicnt) + '_replace_' + str(int(ratio * 100)) + '_' + str(
int(step))
checkin = para_replace(city, defense_name, checkin, ratio, step)
ul_graph, lu_graph = ul_graph_build(checkin, 'locid')
model_name = str(cicnt) + '_locid_replace_' + str(int(
ratio * 100)) + '_' + str(int(step))
walk_len, walk_times = 100, 20 # maximal 100 walk_len, 20 walk_times
print 'walking'
para_ul_random_walk(city, model_name, checkin.uid.unique(), ul_graph, lu_graph,\
walk_len, walk_times)
print 'walk done'
print 'emb training'
emb_train(city, model_name)
print 'emb training done'
