核心语句：

    1. from datetime import datetime
    2. prev_time = datetime.now()
    3. ...
    4. cur_time = datetime.now()
    6. h, remainder = divmod((cur_time - prev_time).seconds, 3600) #divmod的返回值：商和余数
    7. m, s = divmod(remainder, 60)
    8. time_str = "Time %02d:%02d:%02d" % (h, m, s)

    参考下面实例

    1. def train(net, train_data, valid_data, num_epochs, optimizer, criterion):
    2.     prev_time = datetime.now()
    3.     for epoch in range(num_epochs):
    4.         train_loss = 0
    5.         train_acc = 0
    6.         net = net.train()
    7.         for im, label in train_data:
    8.             im = im.to(device)  # (bs, 3, h, w)
    9.             label = label.to(device)  # (bs, h, w)
    10.             # forward
    11.             output = net(im)
    12.             loss = criterion(output, label)
    13.             # backward
    14.             optimizer.zero_grad()
    15.             loss.backward()
    16.             optimizer.step()
    18.             train_loss += loss.item()
    19.             train_acc += get_acc(output, label)
    20.         writer.add_scalar('训练损失值', train_loss, epoch) #记录损失值准备可视化
    21.         cur_time = datetime.now()
    22.         h, remainder = divmod((cur_time - prev_time).seconds, 3600)
    23.         m, s = divmod(remainder, 60)
    24.         time_str = "Time %02d:%02d:%02d" % (h, m, s)
    25.         if valid_data is not None:
    26.             valid_loss = 0
    27.             valid_acc = 0
    28.             net = net.eval()
    29.             for im, label in valid_data:
    30.                 im = im.to(device)  # (bs, 3, h, w)
    31.                 label = label.to(device)  # (bs, h, w)
    32.                 output = net(im)
    33.                 loss = criterion(output, label)
    34.                 valid_loss += loss.item()
    35.                 valid_acc += get_acc(output, label)
    36.             epoch_str = (
    37.                     "Epoch %d. Train Loss: %f, Train Acc: %f, Valid Loss: %f, Valid Acc: %f, "
    38.                     % (epoch, train_loss / len(train_data),
    39.                        train_acc / len(train_data), valid_loss / len(valid_data),
    40.                        valid_acc / len(valid_data)))
    41.         else:
    42.             epoch_str = ("Epoch %d. Train Loss: %f, Train Acc: %f, " %
    43.                          (epoch, train_loss / len(train_data),
    44.                           train_acc / len(train_data)))
    45.         prev_time = cur_time
    46.         print(epoch_str + time_str)