import torch import matplotlib.pyplot as plt import pandas as pd import seaborn as sns from mydataset import MyDataset from torch.utils.data import DataLoader from sklearn.metrics import accuracy_score, confusion_matrix import argparse import torch.nn as nn

def test(model, dataset, criterion, epoch): model.eval() total_batch_num = 0. val_loss = 0 prediction = [] labels = [] feature_list = torch.tensor([]) if torch.cuda.is_available(): feature_list = feature_list.cuda() accuracy_dict = {} for (step, i) in enumerate(dataset): total_batch_num = total_batch_num + 1 batch_x = i['data'] batch_y = i['label'] batch_x = torch.unsqueeze(batch_x, dim=1) batch_x = batch_x.float() if torch.cuda.is_available(): batch_x = batch_x.cuda() batch_y = batch_y.cuda() feature, probs = model(batch_x) batch_label = batch_y.unsqueeze(1).float() feature_label = torch.cat((feature, batch_label), dim=1) feature_list = torch.cat((feature_list, feature_label), dim=0) loss = criterion(probs, batch_y) _, pred = torch.max(probs, dim=1) predi = pred.tolist() label = batch_y.tolist() val_loss += loss.item() prediction.extend(predi) labels.extend(label)

accuracy = accuracy_score(labels, prediction)
C = confusion_matrix(labels, prediction)

if epoch == 68:
    unique_labels = set(labels)
    for label in unique_labels:
        accuracy_dict[label] = accuracy_score([label] * labels.count(label), [label] * labels.count(label))

return accuracy, val_loss / total_batch_num, feature_list, C, accuracy_dict

定义args

parser = argparse.ArgumentParser() rootpath = 'das_data' parser.add_argument('--root2', type=str, default=rootpath + '/test', help='rootpath of valdata') parser.add_argument('--txtpath2', type=str, default=rootpath + '/test/label.txt', help='path pf val_list') parser.add_argument('--batch_size', type=int, default=32, help='batch size for testing') args = parser.parse_args()

将args传递给相关函数

test_dataset = MyDataset(args.root2, args.txtpath2, transform=None) test_loader = DataLoader(dataset=test_dataset, batch_size=args.batch_size, shuffle=True, num_workers=0) model = torch.load('./modelpth/68.pth') test_dataset = MyDataset(args.root2, args.txtpath2, transform=None) test_loader = DataLoader(dataset=test_dataset, batch_size=args.batch_size, shuffle=True, num_workers=0) criterion = nn.CrossEntropyLoss() model.eval() accuracy, _, _, C, accuracy_dict = test(model, test_loader, criterion, epoch=68) fig = plt.figure() ax = fig.add_subplot(111) df = pd.DataFrame(C) sns.heatmap(df, fmt='g', annot=True, annot_kws={'size': 10}, xticklabels=['1', '2', '3', '4', '5', '6'], yticklabels=['1', '2', '3', '4', '5', '6'], cmap='Blues') ax.set_xlabel('Predicted label') ax.set_ylabel('True label') plt.savefig('confusion_matrix.jpg') plt.show()

print('Accuracy: {:.4f}'.format(accuracy)) print('Average Accuracy for each label:') for label, acc in accuracy_dict.items(): print('Label {}: {:.4f}'.format(label, acc)) if 68 in accuracy_dict: # 判断是否有精确率数据 print('Label Accuracy at epoch 68:') for label, accuracy in accuracy_dict.items(): print('Label {}: {:.4f}'.format(label, accuracy)) 输出的各个标签精确度为Average Accuracy for each label: Label 0: 1.0000 Label 1: 1.0000 Label 2: 1.0000 Label 3: 1.0000 Label 4: 1.0000 Label 5: 1.0000 不对为什么内容：根据代码，模型在测试集上的精确度是1，即预测结果与真实标签完全一致。这可能是因为测试集中的样本与训练集中的样本非常相似，导致模型在测试集上表现很好。另外，由于测试集样本较少，可能存在过拟合的情况。建议增加测试集样本数量或者使用交叉验证等方法来评估模型的性能。