我正在使用以下训练函数训练模型：
def _train_epoch(self, epoch):
loss_total = 0.0
front_total = 0.0
back_total = 0.0
corr = []
best_acc = 0.0
for i, (mixture, clean, name, label) in enumerate(self.train_data_loader):
mixture = mixture.to(self.device, dtype=torch.float)
clean = clean.to(self.device, dtype=torch.float)
if i % 10 == 0:
enhanced = self.model(mixture).to(self.device)
front_loss = self.loss_function(clean, enhanced)
model_back.train()
y = model_back(enhanced.float().to(device2))
back_loss = self.loss_function2(y[0], label[0].to(device2))
loss = front_loss + back_loss
print("Iteration %d in epoch%d--> front_loss = %f back_loss = %f loss = %f " % (i, epoch, front_loss.item(), back_loss.item() ,loss.item()), end='\r')
loss_total += loss.item()
front_total += front_loss.item()
back_total += back_loss.item()
loss.backward()
self.optimizer.step()
optimizer_back.step()
self.optimizer.zero_grad()
optimizer_back.zero_grad()
p = torch.argmax(y[0].detach().cpu(), dim=1)
intent_p = p
corr.append((intent_p == label[0]).float())
#torch.cuda.empty_cache()
ac = np.mean(np.hstack(corr))
intent_ac = ac
iter_ac = '\n iteration %d epoch %d -->' %(i, epoch)
print(iter_ac, ac, best_acc)
if intent_ac > best_acc:
improved_acc = 'Current accuracy {}, {}'.format(intent_ac, best_acc)
best_acc = intent_ac
print(improved_acc)
我有一个基本的python问题，但很抱歉我无法解决它，问题是best_acc始终为0并且没有更新，而我正在设置条件以在intent_ac更好时更新其值如图所示，比 best_acc：并不是说 intent_ac 正在改进，因此应该更新 best_acc。 ============== 1757 纪元 ==============
epoch1757 迭代 5990--> front_loss = 0.008438 back_loss = 1.030782 loss = 1.039219 迭代 5999 纪元 1757 --> 0.8016667 0.0
电流精度 0.8016666769981384, 0.0
============== 1758 纪元 ==============
epoch1758 迭代 5990--> front_loss = 0.013248 back_loss = 1.306771 loss = 1.320019 迭代 5999 纪元 1758 --> 0.81 0.0
电流精度 0.8100000023841858, 0.0
============== 1759 纪元 ==============
epoch1759 迭代 5990--> front_loss = 0.008453 back_loss = 1.679812 loss = 1.688265 迭代 5999 纪元 1759 --> 0.81 0.0 电流精度 0.8100000023841858, 0.0