def main():
parser = argparse.ArgumentParser()
parser.add_argument('--method', type=str, default='mean')
parser.add_argument('--level', type=int, default=0)
parser.add_argument('--seed', type=int, default=0)
parser.add_argument('--log_dir', type=str, default='0')
subparsers = parser.add_subparsers()
add_uci_subparser(subparsers)
add_mc_subparser(subparsers)
args = parser.parse_args()
print(args)
seed = args.seed
np.random.seed(seed)
torch.manual_seed(seed)
if args.domain == 'uci':
from uci.uci_data import load_data
data = load_data(args)
elif args.domain == 'mc':
from mc.mc_data import load_data
data = load_data(args)
log_path = './{}/test/{}/{}_{}/'.format(args.domain, args.data,
args.method, args.log_dir)
os.makedirs(log_path)
cmd_input = 'python ' + ' '.join(sys.argv) + '\n'
with open(osp.join(log_path, 'cmd_input.txt'), 'a') as f:
f.write(cmd_input)
baseline_mdi(data, args, log_path)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--domain', type=str, default='uci') # 'uci'
parser.add_argument('--data', type=str, default='housing') # 'pks', 'cancer', 'housing', 'wine'
parser.add_argument('--method', type=str, default='mean')
parser.add_argument('--train_edge', type=float, default=0.7)
parser.add_argument('--train_y', type=float, default=0.7)
parser.add_argument('--log_dir', type=str, default='lry0')
parser.add_argument('--load_dir', type=str, default='0')
parser.add_argument('--seed', type=int, default=0)
args = parser.parse_args()
seed = args.seed
np.random.seed(seed)
torch.manual_seed(seed)
if args.domain == 'uci':
from uci.uci_data import load_data
data = load_data(args)
log_path = './{}/test/{}/{}_{}/'.format(args.domain,args.data,args.method,args.log_dir)
os.mkdir(log_path)
cmd_input = 'python ' + ' '.join(sys.argv) + '\n'
with open(osp.join(log_path, 'cmd_input.txt'), 'a') as f:
f.write(cmd_input)
load_path = './{}/test/{}/{}/'.format(args.domain,args.data,args.load_dir)
linear_regression(data, args, log_path, load_path)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--domain', type=str, default='uci')
parser.add_argument('--data', type=str, default='housing')
parser.add_argument('--train_edge', type=float, default=0.7)
parser.add_argument('--split_sample', type=float, default=0.)
parser.add_argument('--split_by', type=str, default='y') # 'y', 'random'
parser.add_argument('--split_train', action='store_true', default=False)
parser.add_argument('--split_test', action='store_true', default=False)
parser.add_argument('--train_y', type=float, default=0.7)
parser.add_argument(
'--node_mode', type=int,
default=0) # 0: feature onehot, sample all 1; 1: all onehot
parser.add_argument('--method', type=str, default=None)
parser.add_argument('--level', type=int, default=0)
parser.add_argument('--seed', type=int, default=0)
parser.add_argument('--best_level', action='store_true', default=False)
parser.add_argument('--comment', type=str, default='v1')
args = parser.parse_args()
# device is cpu by default
np.random.seed(args.seed)
torch.manual_seed(args.seed)
best_levels = {'mean': 0, 'knn': 2, 'svd': 2, 'mice': 0, 'spectral': 1}
## new
for args.data in [
'concrete', 'energy', 'housing', 'kin8nm', 'naval', 'power',
'protein', 'wine', 'yacht'
]:
for args.method in [
'gnn', 'gnn_mdi', 'mean', 'knn', 'svd', 'mice', 'spectral'
]:
data = load_data(args)
if args.best_level and (not args.method.startswith('gnn')):
args.level = best_levels[args.method]
print("using best level {} for {}".format(
args.level, args.method))
log_path = './uci/y_results/results/{}_{}/{}/{}/'.format(
args.method, args.comment, args.data, args.seed)
if not os.path.isdir(log_path):
os.makedirs(log_path)
if args.method == 'gnn':
load_path = './uci/y_results/results/gnn_{}/{}/{}/'.format(
args.comment, args.data, args.seed)
elif args.method == 'gnn_mdi':
load_path = './uci/mdi_results/results/gnn_mdi_{}/{}/{}/'.format(
args.comment, args.data, args.seed)
else:
load_path = ''
linear_regression(data, args, log_path, load_path)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--domain', type=str, default='uci')
parser.add_argument('--data', type=str, default=None)
parser.add_argument('--train_edge', type=float, default=0.7)
parser.add_argument('--train_y', type=float, default=0.7)
parser.add_argument(
'--node_mode', type=int,
default=0) # 0: feature node onehot, sample node all 1; 1: all onehot
parser.add_argument('--model_types', type=str, default='EGSAGE_EGSAGE')
parser.add_argument(
'--post_hiddens',
type=str,
default=None,
) # default to be 1 hidden of node_dim
parser.add_argument('--concat_states', action='store_true', default=False)
parser.add_argument(
'--norm_embs',
type=str,
default=None,
) # default to be all true
parser.add_argument(
'--aggr',
type=str,
default='mean',
)
parser.add_argument('--node_dim', type=int, default=16)
parser.add_argument('--edge_dim', type=int, default=16)
parser.add_argument('--edge_mode', type=int,
default=1) # 0: use it as weight 1: as input to mlp
parser.add_argument('--gnn_activation', type=str, default='relu')
parser.add_argument('--impute_hiddens', type=str, default='')
parser.add_argument('--impute_activation', type=str, default='relu')
parser.add_argument('--predict_hiddens', type=str, default='')
parser.add_argument('--epochs', type=int, default=20000)
parser.add_argument('--opt', type=str, default='adam')
parser.add_argument('--opt_scheduler', type=str, default='none')
parser.add_argument('--opt_restart', type=int, default=0)
parser.add_argument('--opt_decay_step', type=int, default=1000)
parser.add_argument('--opt_decay_rate', type=float, default=0.9)
parser.add_argument('--dropout', type=float, default=0.)
parser.add_argument('--weight_decay', type=float, default=0.)
parser.add_argument('--lr', type=float, default=0.001)
parser.add_argument('--known', type=float,
default=0.7) # 1 - edge dropout rate
parser.add_argument('--valid', type=float, default=0.) # valid-set ratio
parser.add_argument('--seed', type=int, default=0)
parser.add_argument('--comment', type=str, default='v1')
args = parser.parse_args()
print(args)
# select device
if torch.cuda.is_available():
cuda = auto_select_gpu()
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ['CUDA_VISIBLE_DEVICES'] = str(cuda)
print('Using GPU {}'.format(os.environ['CUDA_VISIBLE_DEVICES']))
device = torch.device('cuda:{}'.format(cuda))
else:
print('Using CPU')
device = torch.device('cpu')
np.random.seed(args.seed)
torch.manual_seed(args.seed)
## new
for args.data in [
'concrete', 'energy', 'housing', 'kin8nm', 'naval', 'power',
'protein', 'wine', 'yacht'
]:
data = load_data(args)
log_path = './uci/y_results/results/gnn_{}/{}/{}/'.format(
args.comment, args.data, args.seed)
if not os.path.isdir(log_path):
os.makedirs(log_path)
# if os.path.isdir(log_path):
# if len(os.listdir(log_path))>0:
# print('Directory exists, skip training')
# continue
# else:
# print('Directory exists, no content, re-initialize training')
# shutil.rmtree(log_path)
train_gnn_y(data, args, log_path, device)
parser = argparse.ArgumentParser()
parser.add_argument('--seed', type=int, default=0)
subparsers = parser.add_subparsers()
add_uci_subparser(subparsers)
add_mc_subparser(subparsers)
args = parser.parse_args()
print(args)
seed = args.seed
np.random.seed(seed)
torch.manual_seed(seed)
if args.domain == 'uci':
from uci.uci_data import load_data
data = load_data(args)
elif args.domain == 'mc':
from mc.mc_data import load_data
data = load_data(args)
# print(data)
print('node dim: ', data.num_node_features, ' edge dim: ', data.edge_attr_dim)
for i, key in enumerate(data.keys):
if torch.is_tensor(data[key]):
print(key, ': ', data[key].shape)
# else:
# print(key,': ',data[key])
if hasattr(args, 'split_sample') and args.split_sample > 0.:
lower_y = data.y[data.lower_y_index]
higher_y = data.y[data.higher_y_index]
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--model_types',
type=str,
default='EGSAGE_EGSAGE_EGSAGE')
parser.add_argument(
'--post_hiddens',
type=str,
default=None,
) # default to be 1 hidden of node_dim
parser.add_argument('--concat_states', action='store_true', default=False)
parser.add_argument(
'--norm_embs',
type=str,
default=None,
) # default to be all true
parser.add_argument(
'--aggr',
type=str,
default='mean',
)
parser.add_argument('--node_dim', type=int, default=64)
parser.add_argument('--edge_dim', type=int, default=64)
parser.add_argument('--edge_mode', type=int,
default=1) # 0: use it as weight; 1: as input to mlp
parser.add_argument('--gnn_activation', type=str, default='relu')
parser.add_argument('--impute_hiddens', type=str, default='64')
parser.add_argument('--impute_activation', type=str, default='relu')
parser.add_argument('--epochs', type=int, default=20000)
parser.add_argument('--opt', type=str, default='adam')
parser.add_argument('--opt_scheduler', type=str, default='none')
parser.add_argument('--opt_restart', type=int, default=0)
parser.add_argument('--opt_decay_step', type=int, default=1000)
parser.add_argument('--opt_decay_rate', type=float, default=0.9)
parser.add_argument('--dropout', type=float, default=0.)
parser.add_argument('--weight_decay', type=float, default=0.)
parser.add_argument('--lr', type=float, default=0.001)
parser.add_argument('--known', type=float,
default=0.7) # 1 - edge dropout rate
parser.add_argument('--auto_known', action='store_true', default=False)
parser.add_argument(
'--loss_mode', type=int, default=0
) # 0: loss on all train edge, 1: loss only on unknown train edge
parser.add_argument('--valid', type=float, default=0.) # valid-set ratio
parser.add_argument('--seed', type=int, default=0)
parser.add_argument('--log_dir', type=str, default='0')
parser.add_argument('--save_model', action='store_true', default=False)
parser.add_argument('--save_prediction',
action='store_true',
default=False)
parser.add_argument('--transfer_dir', type=str, default=None)
parser.add_argument('--transfer_extra', type=str, default='')
parser.add_argument('--mode', type=str, default='train') # debug
subparsers = parser.add_subparsers()
add_uci_subparser(subparsers)
add_mc_subparser(subparsers)
args = parser.parse_args()
print(args)
# select device
if torch.cuda.is_available():
cuda = auto_select_gpu()
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ['CUDA_VISIBLE_DEVICES'] = str(cuda)
print('Using GPU {}'.format(os.environ['CUDA_VISIBLE_DEVICES']))
device = torch.device('cuda:{}'.format(cuda))
else:
print('Using CPU')
device = torch.device('cpu')
seed = args.seed
np.random.seed(seed)
torch.manual_seed(seed)
if args.domain == 'uci':
from uci.uci_data import load_data
data = load_data(args)
elif args.domain == 'mc':
from mc.mc_data import load_data
data = load_data(args)
log_path = './{}/test/{}/{}/'.format(args.domain, args.data, args.log_dir)
os.makedirs(log_path)
cmd_input = 'python ' + ' '.join(sys.argv) + '\n'
with open(osp.join(log_path, 'cmd_input.txt'), 'a') as f:
f.write(cmd_input)
train_gnn_mdi(data, args, log_path, device)