test_acc, test_acc_top5, test_loss = validate(val_loader, model, criterion, opt)

在PyTorch中，`validate()` 函数通常用于评估模型在验证集上的性能。这里给出了`validate()`函数可能的调用方式以及涉及的关键参数： 1. `test_acc`: 测试准确率[^1]。这通常是通过计算预测类别与实际标签相符的情况来确定的。在每个批次的验证数据上，模型会应用`model(x)`得到预测结果，然后与`criterion`计算损失。最后，根据正确的分类数量除以总样本数，得出整个验证集的准确率。 ```python with torch.no_grad(): # 关闭梯度计算以节省内存 correct = 0 total = 0 for images, labels in val_loader: outputs = model(images) _, predicted = torch.max(outputs.data, 1) total += labels.size(0) correct += (predicted == labels).sum().item() test_acc = correct / total ``` 2. `test_acc_top5`: 测试前五名准确率。这表示对于每个样本，如果模型前五个预测中最少有一个与真实标签相同，则认为该样本被正确预测。实现方法可能包括对输出概率最高的前五个类进行判断。 ```python top5_correct = 0 for images, labels in val_loader: ... top5_pred = torch.topk(outputs, 5, dim=1)[1] top5_correct += (top5_pred[:, :labels.size(1)] == labels.unsqueeze(1)).any(dim=1).sum().item() test_acc_top5 = top5_correct / total ``` 3. `test_loss`: 测试损失。这是通过`criterion`函数计算模型输出与真实标签之间的差异，如交叉熵损失。在每个批次之后累加损失，最后除以总批次数得到平均损失。 ```python test_loss = 0. for images, labels in val_loader: ... loss = criterion(outputs, labels) test_loss += loss.item() test_loss /= len(val_loader) ``` 完整代码示例： ```python def validate(val_loader, model, criterion, opt=None): test_loss = 0. correct = 0 top5_correct = 0 with torch.no_grad(): for images, labels in val_loader: outputs = model(images) loss = criterion(outputs, labels) _, predicted = torch.max(outputs.data, 1) correct += (predicted == labels).sum().item() top5_pred = torch.topk(outputs, 5, dim=1)[1] top5_correct += (top5_pred[:, :labels.size(1)] == labels.unsqueeze(1)).any(dim=1).sum().item() test_loss += loss.item() test_loss /= len(val_loader) test_acc = correct / total test_acc_top5 = top5_correct / total if opt and opt.verbose: print(f"Test Loss: {test_loss:.4f}, Test Acc: {test_acc * 100:.2f}%, Top5 Acc: {test_acc_top5 * 100:.2f}%") return test_loss, test_acc, test_acc_top5 ```