Question

我有以下分类模型。我有训练集和测试集。我在训练集上训练它，输入是 3400 向量，输出是 3 个类 (0,1,2) 之间的一个类。我将模型保存为以下代码。现在我想应用 10 折交叉验证来评估测试集上保存的模型。你能告诉我怎么做吗，因为我以前从未使用过 10 交叉验证。

training_set = Dataset("train_data.txt","train_target.txt")
training_generator = torch.utils.data.DataLoader(training_set, **params)
testing_set = Dataset("test_data.txt","testtarget.txt")
testing_generator = torch.utils.data.DataLoader(testing_set, **params)
    for i, (seq_batch, stat_batch) in enumerate(training_generator):
        seq_batch, stat_batch = seq_batch.to(device), stat_batch.to(device)
        optimizer.zero_grad()
        #print(seq_batch.shape,stat_batch.shape)
        # Model computation
        seq_batch = seq_batch.unsqueeze(-1)
        outputs = model(seq_batch)
        if CUDA:
            loss = criterion(outputs, stat_batch)
        loss.backward()
        optimizer.step()
        # print statistics
        running_loss += loss.item()
        epoch_loss += loss.item()*outputs.shape[0]
        if i % 2000 == 1999:  # print every 2000 mini-batches
            print('[%d, %5d] loss: %.3f' %
                  (epoch + 1, i + 1, running_loss / 2000),"acc",(outputs.argmax(1) == stat_batch).float().mean())
            running_loss = 0.0
        sum_acc += (outputs.argmax(1) == stat_batch).float().sum()

    print("epoch" , epoch+1, "acc", sum_acc/len(training_set),"loss", epoch_loss/len(training_set))
    loss_values.append(epoch_loss/len(training_set))
    if epoch % 20 == 0:
        torch.save(model.state_dict(), path + name_file + "model_epoch_i_" + str(epoch) + ".cnn")

Answer 1

这个话题可能对你有用。答案之一包含自定义 CV 函数： k-fold cross validation using DataLoaders in PyTorch

rm -rf node_modules && npm install

10 折交叉验证评估

1 个答案: