多输出模型评估与损失加权

PyTorch 深度学习进阶

Michal Oleszak

Machine Learning Engineer

模型评估

acc_alpha = Accuracy(
    task="multiclass", num_classes=30
)
acc_char = Accuracy(
    task="multiclass", num_classes=964
)


net.eval()
with torch.no_grad():
    for images, labels_alpha, labels_char \
    in dataloader_test:
        out_alpha, out_char = net(images)

        _, pred_alpha = torch.max(out_alpha, 1)
        _, pred_char = torch.max(out_char, 1)

        acc_alpha(pred_alpha, labels_alpha)
        acc_char(pred_char, labels_char)

为每个输出设置度量
遍历测试加载器并获取输出
计算各输出的预测
更新准确率度量
计算最终准确率

print(f"Alphabet: {acc_alpha.compute()}")
print(f"Character: {acc_char.compute()}")

Alphabet: 0.3166305720806122
Character: 0.24064336717128754

多输出训练循环回顾

for epoch in range(10):
    for images, labels_alpha, labels_char \
    in dataloader_train:
        optimizer.zero_grad()
        outputs_alpha, outputs_char = net(images)
        loss_alpha = criterion(
          outputs_alpha, labels_alpha
        )
        loss_char = criterion(
          outputs_char, labels_char
        )
        loss = loss_alpha + loss_char
        loss.backward()
        optimizer.step()

两个损失：字母表与字符
最终损失为二者之和：loss = loss_alpha + loss_char
两个分类任务同等重要

调整任务重要性

字符分类比字母表分类重要 2 倍

方法 1：将更重要的损失放大 2 倍
```
loss = loss_alpha + loss_char * 2
```

方法 2：分配权重且权重和为 1

loss = 0.33 * loss_alpha + 0.67 * loss_char

警示：损失量级不一致

在加权求和前，损失需在同一量级
示例任务：
- 预测房价 -> MSE 损失
- 预测质量：低/中/高 -> CrossEntropy 损失
CrossEntropy 通常为个位数
MSE 可达上万
模型会忽视质量评估任务

解决：先归一化两者再加权求和

loss_price = loss_price / torch.max(loss_price)
loss_quality = loss_quality / torch.max(loss_quality)
loss = 0.7 * loss_price + 0.3 * loss_quality

Ayo berlatih!

PyTorch 深度学习进阶