l_tau = ['tau_dr', 'tau_ols', 'tau_ols_ps', 'tau_resid']
output = 'results/2019-10-30_' + exp_name + '.csv'
l_scores = []
for args['citcio'] in [False, True]:
for args['model'] in ["dlvm", "lrmf"]:
for args['seed'] in range_seed:
for args['prop_miss'] in range_prop_miss:
for args['n'] in range_n:
for args['p'] in range_p:
t0 = time.time()
score = exp_mi(**args)
args['time'] = int(time.time() - t0)
l_scores.append(
np.concatenate((list(args.values()), score)))
print('exp with ', args)
print('........... DONE')
print('in ', int(args["time"]), ' s \n\n')
score_data = pd.DataFrame(l_scores,
columns=list(args.keys()) +
l_tau)
score_data.to_csv(output + '_temp')
print('saving ' + exp_name + 'at: ' + output)
score_data.to_csv(output)
print('*' * 20)
print('Exp: ' + exp_name + ' succesfully ended.')
print('*' * 20)
args['m'] = 10
print('starting exp: ' + exp_name)
l_tau = ['tau_dr', 'tau_ols', 'tau_ols_ps']
output = 'results/2019-10-24_'+exp_name+'.csv'
l_scores = []
for args['citcio'] in [False, True]:
for args['model'] in ["dlvm","lrmf"]:
for args['seed'] in range_seed:
for args['prop_miss'] in range_prop_miss:
for args['n'] in range_n:
for args['p'] in range_p:
t0 = time.time()
score = exp_mi(**args)
args['time'] = int(time.time() - t0)
l_scores.append(np.concatenate((list(args.values()),score)))
print('exp with ', args)
print('........... DONE')
print('in ', int(args["time"]) , ' s \n\n')
score_data = pd.DataFrame(l_scores, columns=list(args.keys()) + l_tau)
score_data.to_csv(output + '_temp')
print('saving ' +exp_name + 'at: ' + output)
score_data.to_csv(output)
print('*'*20)
print('Exp: '+ exp_name+' succesfully ended.')
print('*'*20)
l_scores = []
for args['seed'] in range_seed:
for args['citcio'] in [False, True]:
for args['model'] in ["dlvm", "lrmf"]:
for args['n'] in range_n:
for args['sig_prior'] in range_sig_prior:
for args['n_epochs'] in range_n_epochs:
for args['prop_miss'] in range_prop_miss:
for args['p'] in range_p:
t0 = time.time()
score = exp_miwae(**args)
args['time'] = int(time.time() - t0)
l_scores.append(
np.concatenate(
(list(args.values()), score)))
print('exp with ', args)
print('........... DONE')
print('in ', int(args["time"]), ' s \n\n')
score_data = pd.DataFrame(
l_scores, columns=list(args.keys()) + l_tau)
score_data.to_csv(output + '_temp')
print('saving ' + exp_name + 'at: ' + output)
score_data.to_csv(output)
print('*' * 20)
print('Exp: ' + exp_name + ' succesfully ended.')
print('*' * 20)
type=str,
default=default_args['subtask_id'],
required=False,
choices=['A', 'B'],
help='subtask A or B (default: {})'.format(
default_args['subtask_id']))
parser.add_argument('--no-cuda',
action='store_true',
default=False,
help='disables CUDA training')
# parser.add_argument('--dry-run', action='store_true', default=False,
# help='quickly check a single pass')
# parser.add_argument('--seed', type=int, default=1, metavar='S',
# help='random seed (default: 1)')
# parser.add_argument('--log-interval', type=int, default=10, metavar='N',
# help='how many batches to wait before logging training status')
# parser.add_argument('--save-model', action='store_true', default=False,
# help='For Saving the current Model')
args = parser.parse_args() # 获取用户输入的参数
# parser.print_help()
for key in default_args.keys():
# 将输入的参数更新至 default_args
if hasattr(args, key):
default_args[key] = getattr(args, key)
default_args['use_cuda'] = not args.no_cuda and torch.cuda.is_available()
default_args['device'] = torch.device(
'cuda:0' if torch.cuda.is_available() else 'cpu')
pprint.pprint(default_args)
tau = []
t0 = time.time()
for m in range_m:
tau.append(exp_mi(X_miss, w, y, m=m))
args['time'] = int(time.time() - t0)
for i in range(len(tau)):
l_scores.append(np.concatenate((['MI'], list(args.values()), [m], [None]*6, tau[i])))
# MF
t0 = time.time()
tau = exp_mf(X_miss, w, y)
args['time'] = int(time.time() - t0)
l_scores.append(np.concatenate((['MF'], list(args.values()), [None], [tau[-1]], [None]*5, tau[:-1])))
score_data = pd.DataFrame(l_scores, columns=['Method'] + list(args.keys()) + l_method_params + l_tau)
print(score_data)
# MDC
range_d_miwae = [args['d']+x for x in range_d_offset]
t0 = time.time()
tau, params = exp_mdc(X_miss, w, y,
range_d_miwae = range_d_miwae,
range_sig_prior = range_sig_prior,
range_num_samples_zmul = range_num_samples_zmul,
range_learning_rate = range_learning_rate,
range_n_epochs = range_n_epochs)
args['time'] = int(time.time() - t0)
for i in range(len(tau['MDC.process'])):
l_scores.append(np.concatenate((['MDC.process'], list(args.values()), [None]*2, params[i], tau['MDC.process'][i])))
for i in range(len(tau['MDC.mi'])):
def build_parse():
parser = argparse.ArgumentParser(
description='ECNU-SenseMaker single model')
parser.add_argument(
'--batch-size',
type=int,
default=default_args['batch_size'],
metavar='N',
help='input batch size for training (default: {})'.format(
default_args['batch_size']))
parser.add_argument(
'--test-batch-size',
type=int,
default=default_args['test_batch_size'],
metavar='N',
help='input batch size for testing (default: {})'.format(
default_args['test_batch_size']))
parser.add_argument('--epochs',
type=int,
default=default_args['epochs'],
metavar='N',
help='number of epochs to train (default: {})'.format(
default_args['epochs']))
parser.add_argument(
'--fine-tune-epochs',
type=int,
default=default_args['fine_tune_epochs'],
metavar='N',
help='number of fine-tune epochs to train (default: {})'.format(
default_args['fine_tune_epochs']))
parser.add_argument('--lr',
type=float,
default=default_args['lr'],
metavar='LR',
help='learning rate (default: {})'.format(
default_args['lr']))
parser.add_argument('--fine-tune-lr',
type=float,
default=default_args['fine_tune_lr'],
metavar='LR',
help='fine-tune learning rate (default: {})'.format(
default_args['fine_tune_lr']))
parser.add_argument('--adam-epsilon',
type=float,
default=default_args['adam_epsilon'],
metavar='M',
help='Adam epsilon (default: {})'.format(
default_args['adam_epsilon']))
parser.add_argument('--max-seq-length',
type=int,
default=default_args['max_seq_length'],
metavar='N',
help='max length of sentences (default: {})'.format(
default_args['max_seq_length']))
parser.add_argument('--subtask-id',
type=str,
default=default_args['subtask_id'],
required=False,
choices=['A', 'B'],
help='subtask A or B (default: {})'.format(
default_args['subtask_id']))
parser.add_argument('--with-lm',
action='store_true',
default=False,
help='Add Internal Sharing Mechanism (LM)')
parser.add_argument(
'--with-kegat',
action='store_true',
default=False,
help='Add Knowledge-enhanced Graph Attention Network (KEGAT)')
parser.add_argument('--with-kemb',
action='store_true',
default=False,
help='Add Knowledge-enhanced Embedding (KEmb)')
parser.add_argument('--no-cuda',
action='store_true',
default=False,
help='disables CUDA training')
# parser.add_argument('--dry-run', action='store_true', default=False,
# help='quickly check a single pass')
parser.add_argument('--seed',
type=int,
default=1,
metavar='S',
help='random seed (default: 1)')
# parser.add_argument('--save-model', action='store_true', default=False,
# help='For Saving the current Model')
args = parser.parse_args() # 获取用户输入的参数
torch.manual_seed(args.seed)
for key in default_args.keys():
# 将输入的参数更新至 default_args
if hasattr(args, key):
default_args[key] = getattr(args, key)
default_args['use_cuda'] = not args.no_cuda and torch.cuda.is_available()
default_args['device'] = torch.device(
'cuda:0' if default_args['use_cuda'] else 'cpu')
default_args['with_lm'] = args.with_lm
default_args['with_kegat'] = args.with_kegat
default_args['with_kemb'] = args.with_kemb
return args
