def save_test_submission(self, input_dict, dir_path):
assert 'y_pred' in input_dict
y_pred = input_dict['y_pred']
assert y_pred.shape == (146818, )
if isinstance(y_pred, torch.Tensor):
y_pred = y_pred.cpu().numpy()
y_pred = y_pred.astype(np.short)
makedirs(dir_path)
filename = osp.join(dir_path, 'y_pred_mag240m')
np.savez_compressed(filename, y_pred=y_pred)
def save_test_submission(self, input_dict, dir_path):
assert 'h,r->t' in input_dict
assert 't_pred_top10' in input_dict['h,r->t']
t_pred_top10 = input_dict['h,r->t']['t_pred_top10']
assert t_pred_top10.shape == (1359303, 10) and (0 <= t_pred_top10).all() and (t_pred_top10 < 1001).all()
if isinstance(t_pred_top10, torch.Tensor):
t_pred_top10 = t_pred_top10.cpu().numpy()
t_pred_top10 = t_pred_top10.astype(np.int16)
makedirs(dir_path)
filename = osp.join(dir_path, 't_pred_wikikg90m')
np.savez_compressed(filename, t_pred_top10=t_pred_top10)
def save_test_submission(self, input_dict, dir_path):
assert 'h,r->t' in input_dict
assert 't_pred_top10' in input_dict['h,r->t']
t_pred_top10 = input_dict['h,r->t']['t_pred_top10']
assert t_pred_top10.shape == (1359303, 10) and (
0 <= t_pred_top10).all() and (t_pred_top10 < 1001).all()
if isinstance(t_pred_top10, torch.Tensor):
t_pred_top10 = t_pred_top10.cpu().numpy()
for i in range(len(t_pred_top10)):
assert len(pd.unique(t_pred_top10[i])) == len(
t_pred_top10[i]
), 'Found duplicated tail prediction for some triplets!'
t_pred_top10 = t_pred_top10.astype(np.int16)
makedirs(dir_path)
filename = osp.join(dir_path, 't_pred_wikikg90m')
np.savez_compressed(filename, t_pred_top10=t_pred_top10)
def save_test_submission(self, input_dict: Dict, dir_path: str, mode: str):
assert 'y_pred' in input_dict
assert mode in ['test-whole', 'test-dev', 'test-challenge']
y_pred = input_dict['y_pred']
if mode == 'test-whole':
assert y_pred.shape == (146818, )
filename = osp.join(dir_path, 'y_pred_mag240m')
elif mode == 'test-dev':
assert y_pred.shape == (88092, )
filename = osp.join(dir_path, 'y_pred_mag240m_test-dev')
elif mode == 'test-challenge':
assert y_pred.shape == (58726, )
filename = osp.join(dir_path, 'y_pred_mag240m_test-challenge')
makedirs(dir_path)
if isinstance(y_pred, torch.Tensor):
y_pred = y_pred.cpu().numpy()
y_pred = y_pred.astype(np.short)
np.savez_compressed(filename, y_pred=y_pred)
def save_test_submission(self, input_dict: Dict, dir_path: str, mode: str):
assert 'h,r->t' in input_dict
assert 't_pred_top10' in input_dict['h,r->t']
assert mode in ['test-dev', 'test-challenge']
t_pred_top10 = input_dict['h,r->t']['t_pred_top10']
for i in range(len(t_pred_top10)):
assert len(pd.unique(t_pred_top10[i])) == len(t_pred_top10[i]), 'Found duplicated tail prediction for some triplets!'
if mode == 'test-dev':
assert t_pred_top10.shape == (15000, 10)
filename = osp.join(dir_path, 't_pred_wikikg90m-v2_test-dev')
elif mode == 'test-challenge':
assert t_pred_top10.shape == (10000, 10)
filename = osp.join(dir_path, 't_pred_wikikg90m-v2_test-challenge')
makedirs(dir_path)
if isinstance(t_pred_top10, torch.Tensor):
t_pred_top10 = t_pred_top10.cpu().numpy()
t_pred_top10 = t_pred_top10.astype(np.int32)
np.savez_compressed(filename, t_pred_top10=t_pred_top10)
print("Reading training node features...", end=" ", flush=True)
x_train = dataset.paper_feat[train_idx]
x_train = torch.from_numpy(x_train).to(torch.float).to(device)
print(f"Done! [{time.perf_counter() - t:.2f}s]")
t = time.perf_counter()
print("Reading validation node features...", end=" ", flush=True)
x_valid = dataset.paper_feat[valid_idx]
x_valid = torch.from_numpy(x_valid).to(torch.float).to(device)
print(f"Done! [{time.perf_counter() - t:.2f}s]")
y_train = torch.from_numpy(dataset.paper_label[train_idx])
y_train = y_train.to(device, torch.long)
y_valid = torch.from_numpy(dataset.paper_label[valid_idx])
y_valid = y_valid.to(device, torch.long)
makedirs("results/cs")
model = MLP(
dataset.num_paper_features,
args.hidden_channels,
dataset.num_classes,
args.num_layers,
args.dropout,
not args.no_batch_norm,
args.relu_last,
).to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr)
num_params = sum([p.numel() for p in model.parameters()])
print(f"#Params: {num_params}")
best_valid_acc = 0
for epoch in range(1, args.epochs + 1):
print('Reading training node features...', end=' ', flush=True)
x_train = dataset.paper_feat[train_idx]
x_train = torch.from_numpy(x_train).to(torch.float).to(device)
print(f'Done! [{time.perf_counter() - t:.2f}s]')
t = time.perf_counter()
print('Reading validation node features...', end=' ', flush=True)
x_valid = dataset.paper_feat[valid_idx]
x_valid = torch.from_numpy(x_valid).to(torch.float).to(device)
print(f'Done! [{time.perf_counter() - t:.2f}s]')
y_train = torch.from_numpy(dataset.paper_label[train_idx])
y_train = y_train.to(device, torch.long)
y_valid = torch.from_numpy(dataset.paper_label[valid_idx])
y_valid = y_valid.to(device, torch.long)
makedirs('results/cs')
model = MLP(dataset.num_paper_features, args.hidden_channels,
dataset.num_classes, args.num_layers, args.dropout,
not args.no_batch_norm, args.relu_last).to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr)
num_params = sum([p.numel() for p in model.parameters()])
print(f'#Params: {num_params}')
best_valid_acc = 0
for epoch in range(1, args.epochs + 1):
loss = train(model, x_train, y_train, args.batch_size, optimizer)
train_acc = test(model, x_train, y_train, evaluator)
valid_acc = test(model, x_valid, y_valid, evaluator)
if valid_acc > best_valid_acc:
best_valid_acc = valid_acc
torch.save(model.state_dict(), 'results/cs/model.pt')