def get_aucs(data_list):
START = time.time()
FILE = 'output/{}-{}-epochs-{}-avg.csv'.format(SLURM_JOB_ID,
FLAGS.iterations,
NUM_TO_AVG)
settings = {
'data_name': None,
'alpha': None,
'iterations': FLAGS.iterations,
'model': model
}
with open(FILE, 'a') as f:
w = csv.writer(f)
for dataname, alpha in itertools.product(data_list, alphas):
settings['data_name'] = dataname
settings['alpha'] = alpha
for _ in range(NUM_TO_AVG):
runner = AnomalyDetectionRunner(settings)
exc_info = None
try:
r = runner.erun()
r.append(T(START))
w.writerow(r)
f.flush()
except Exception as e:
exc_info = sys.exc_info()
finally:
if exc_info:
traceback.print_exception(*exc_info)
del exc_info
flags.DEFINE_float('weight_decay', 0.,
'Weight for L2 loss on embedding matrix.')
flags.DEFINE_float('dropout', 0., 'Dropout rate (1 - keep probability).')
flags.DEFINE_integer('features', 1, 'Whether to use features (1) or not (0).')
flags.DEFINE_integer('seed', 50, 'seed for fixing the results.')
flags.DEFINE_integer('iterations', 300, 'number of iterations.')
flags.DEFINE_float('alpha', 0.8, 'balance parameter')
'''
We did not set any seed when we conducted the experiments described in the paper;
We set a seed here to steadily reveal better performance of ARGA
'''
seed = 7
np.random.seed(seed)
tf.set_random_seed(seed)
data_list = ['twitter', 'BlogCatalog', 'Amazon']
dataname = data_list[0]
model = 'gcn_ae'
task = 'anomaly_detection'
settings = {
'data_name': dataname,
'iterations': FLAGS.iterations,
'model': model
}
runner = None
if task == 'anomaly_detection':
runner = AnomalyDetectionRunner(settings)
runner.erun()
settings = {
'data_name': dataset_str,
'iterations': FLAGS.iterations,
'model': model,
'decoder_act': decoder_act
}
results_dir = os.path.sep.join(
['results', dataset_str, task, model])
log_dir = os.path.sep.join([
'logs', dataset_str, task, model,
'{}_{}_{}'.format(eta, theta, alpha)
])
if not os.path.exists(results_dir):
os.makedirs(results_dir)
if not os.path.exists(log_dir):
os.makedirs(log_dir)
file2print = '{}/{}_{}_{}_{}_{}.json'.format(
results_dir, dataset_str, eta, theta, alpha, embed_dim)
runner = None
if task == 'anomaly_detection':
runner = AnomalyDetectionRunner(settings, k)
writer = SummaryWriter(log_dir)
runner.erun(writer)