def train_model_time(dataset_str, depth=None, res_connection=False):
A_norm, features, labels, _, _, _ = load_data(dataset_str)
idx_train = np.arange(len(features))
idx_train = torch.LongTensor(idx_train)
def train(model, idx_train):
model.train()
output = model(features)
loss_train = loss(output, labels, idx_train)
acc_train = accuracy(output, labels, idx_train)
optimizer.zero_grad()
loss_train.backward()
optimizer.step()
return loss_train.item(), acc_train.item()
def evaluate(model, idx):
model.eval()
output = model(features)
loss_ = loss(output, labels, idx)
acc_ = accuracy(output, labels, idx)
return loss_.item(), acc_.item()
model = MutipleGCN(adj=A_norm,
ngcu=depth,
nfeat=features.shape[1],
nhid=hidden,
nclass=labels.max().item() + 1,
dropout=dropout,
res_connection=res_connection).to(device=device)
optimizer = torch.optim.Adam(model.parameters(),
lr=lr,
weight_decay=weight_decay)
t_start = time.time()
for i in range(10):
loss_train, acc_train = train(model, idx_train)
training_time = (time.time() - t_start) / 10
t_start = time.time()
for i in range(10):
loss_val, acc_val = evaluate(model, idx_train)
inference_time = (time.time() - t_start) / 10
return training_time, inference_time
self.gc1 = GCN(nfeat, nhid)
self.gc2 = GCN(nhid, nclass)
self.dropout = dropout
def forward(self, x, adj):
x = F.relu(self.gc1(x, adj))
x = F.dropout(x, self.dropout, training=self.training)
x = self.gc2(x, adj)
return F.log_softmax(x, dim=1)
# Load data
data_set = 'cora'
#data_set = 'citeseer'
adj, features, labels, idx_train, idx_val, idx_test = load_data(data_set)
# Model and optimizer
model = GCN_Net(nfeat=features.shape[1],
nhid=args.hidden,
nclass=labels.max().item() + 1,
dropout=args.dropout)
optimizer = optim.Adam(model.parameters(),
lr=args.lr,
weight_decay=args.weight_decay)
def train(epoch):
t = time.time()
model.train()
def train_model_depth(dataset_str, depth=None, res_connection=False):
G, A_norm, features, labels, idx_train, idx_test, idx_val = load_data(
dataset_str)
def train(idx_train):
model.train()
output = model(features)
loss_train = loss(output, labels, idx_train)
acc_train = accuracy(output, labels, idx_train)
optimizer.zero_grad()
loss_train.backward()
optimizer.step()
return loss_train.item(), acc_train.item()
def evaluate(idx):
model.eval()
output = model(features)
loss_ = loss(output, labels, idx)
acc_ = accuracy(output, labels, idx)
return loss_.item(), acc_.item()
nfolds = 5
kf = KFold(n_splits=nfolds)
index_list = np.arange(len(features))
total_train_acc = []
total_test_acc = []
t = 0
for idx_train, idx_test in kf.split(index_list):
model = MutipleGCN(adj=A_norm,
ngcu=depth,
nfeat=features.shape[1],
nhid=hidden,
nclass=labels.max().item() + 1,
dropout=dropout,
res_connection=res_connection).to(device=device)
optimizer = torch.optim.Adam(model.parameters(),
lr=lr,
weight_decay=weight_decay)
idx_train = torch.LongTensor(idx_train)
idx_test = torch.LongTensor(idx_test)
# Train model
t_total = time.time()
epoch_no_improvement = 0
min_val_loss = float('inf')
for epoch in tqdm(range(epochs)):
loss_train, acc_train = train(idx_train)
loss_train, acc_train = evaluate(idx_train)
loss_val, acc_val = evaluate(idx_test)
t += 1
total_train_acc.append(acc_train)
total_test_acc.append(acc_val)
return total_train_acc, total_test_acc