model = CharacterLevelCNN(ops, embedding)
## 3. 声明loss,metric,optimizer
loss = CrossEntropyLoss
metric = AccuracyMetric
#optimizer= SGD([param for param in model.parameters() if param.requires_grad==True], lr=ops.lr)
optimizer = SGD(
[param for param in model.parameters() if param.requires_grad == True],
lr=ops.lr,
momentum=0.9,
weight_decay=ops.weight_decay)
callbacks = []
# callbacks.append(LRScheduler(CosineAnnealingLR(optimizer, 5)))
callbacks.append(
LRScheduler(
LambdaLR(
optimizer, lambda epoch: ops.lr
if epoch < ops.train_epoch * 0.8 else ops.lr * 0.1)))
## 4.定义train方法
def train(model, datainfo, loss, metrics, optimizer, num_epochs=100):
trainer = Trainer(datainfo.datasets['train'],
model,
optimizer=optimizer,
loss=loss(target='target'),
batch_size=ops.batch_size,
metrics=[metrics(target='target')],
dev_data=datainfo.datasets['test'],
device=[0, 1, 2],
check_code_level=-1,
n_epochs=num_epochs)
print(datainfo)
print(datainfo.datasets['train'][0])
model = DPCNN(init_embed=embedding, num_cls=len(datainfo.vocabs[C.TARGET]),
embed_dropout=ops.embed_dropout, cls_dropout=ops.cls_dropout)
print(model)
# 3. 声明loss,metric,optimizer
loss = CrossEntropyLoss(pred=C.OUTPUT, target=C.TARGET)
metric = AccuracyMetric(pred=C.OUTPUT, target=C.TARGET)
optimizer = SGD([param for param in model.parameters() if param.requires_grad == True],
lr=ops.lr, momentum=0.9, weight_decay=ops.weight_decay)
callbacks = []
callbacks.append(LRScheduler(CosineAnnealingLR(optimizer, 5)))
# callbacks.append(
# LRScheduler(LambdaLR(optimizer, lambda epoch: ops.lr if epoch <
# ops.train_epoch * 0.8 else ops.lr * 0.1))
# )
# callbacks.append(
# FitlogCallback(data=datainfo.datasets, verbose=1)
# )
device = 'cuda:0' if torch.cuda.is_available() else 'cpu'
print(device)
# 4.定义train方法
trainer = Trainer(datainfo.datasets['train'], model, optimizer=optimizer, loss=loss,