| 20 | """ |
| 21 | |
| 22 | def plot_roc( |
| 23 | model_name, |
| 24 | label, |
| 25 | predict, |
| 26 | color, |
| 27 | linestyle=":", |
| 28 | linewidth=2, |
| 29 | ): |
| 30 | y_test = torch.load(label) |
| 31 | y_test = y_test.numpy().astype(np.int32) |
| 32 | try: |
| 33 | y_score = torch.load(predict) |
| 34 | y_score = y_score.numpy().astype(np.float64) |
| 35 | except: |
| 36 | y_score = np.load(predict) |
| 37 | y_score = y_score[0:, ].astype(np.float64) |
| 38 | n_classes = 7 |
| 39 | |
| 40 | # 计算每一类的ROC |
| 41 | fpr = dict() |
| 42 | tpr = dict() |
| 43 | roc_auc = dict() |
| 44 | for i in range(n_classes): |
| 45 | fpr[i], tpr[i], _ = roc_curve(y_test[:, i], y_score[:, i]) |
| 46 | roc_auc[i] = auc(fpr[i], tpr[i]) |
| 47 | |
| 48 | # Compute micro-average ROC curve and ROC area（方法二） |
| 49 | fpr["micro"], tpr["micro"], _ = roc_curve(y_test.ravel(), y_score.ravel()) |
| 50 | roc_auc["micro"] = auc(fpr["micro"], tpr["micro"]) |
| 51 | |
| 52 | # Compute macro-average ROC curve and ROC area（方法一） |
| 53 | # First aggregate all false positive rates |
| 54 | all_fpr = np.unique(np.concatenate([fpr[i] for i in range(n_classes)])) |
| 55 | # Then interpolate all ROC curves at this points |
| 56 | mean_tpr = np.zeros_like(all_fpr) |
| 57 | for i in range(n_classes): |
| 58 | mean_tpr += interp(all_fpr, fpr[i], tpr[i]) |
| 59 | # Finally average it and compute AUC |
| 60 | mean_tpr /= n_classes |
| 61 | fpr["macro"] = all_fpr |
| 62 | tpr["macro"] = mean_tpr |
| 63 | roc_auc["macro"] = auc(fpr["macro"], tpr["macro"]) |
| 64 | |
| 65 | # Plot all ROC curves |
| 66 | lw = 2 |
| 67 | # plt.plot(fpr["micro"], tpr["micro"], |
| 68 | # label='micro-average ROC curve of SwimTransformer(area = {0:0.2f})' |
| 69 | # ''.format(roc_auc["micro"]), |
| 70 | # color='red', linestyle='-', linewidth=2) |
| 71 | plt.plot(fpr["macro"], tpr["macro"], |
| 72 | label='{0} (AUC = {1:.3f})'.format(model_name, roc_auc["macro"]), |
| 73 | color=color, linestyle=linestyle, linewidth=linewidth) |
| 74 | |
| 75 | |
| 76 | ##ours |