AUC 1，但准確度 <100%

Question

當測試二元分類器時，我得到 83% 的准確度（當閾值設置為 0.5 時），但是當我鍛煉 ROC 和 AUC 時，我得到的 AUC 值為 1，我認為這是不正確的，因為在這種情況下我應該得到精度為 100？

我有以下數據（例如前 5 點）：

真實標簽： true_list = [1. 1. 1. 1. 1.] true_list = [1. 1. 1. 1. 1.]

閾值預測pred_list = [0. 0. 1. 1. 1.] pred_list = [0. 0. 1. 1. 1.]

原始乙狀結腸激活 output pred_list_raw = [0.23929074 0.34403923 0.61575216 0.72756131 0.69771088]

用於從 model 生成數據的代碼是：

output_raw = Net(images)
output = torch.sigmoid(output_raw)
pred_tag = torch.round(output)

[pred_list.append(pred_tag[i]) for i in range(len(pred_tag.squeeze().cpu().numpy()))]
            
[pred_list_raw.append(output[i]) for i in range(len(output.squeeze().cpu().numpy()))]

ROC 和 AUC 值是使用 sklearn 指標和以下代碼計算的：

fpr, tpr, _ = metrics.roc_curve(true_list,  pred_list_raw)
auc = metrics.roc_auc_score(true_list, pred_list_raw)

准確性和 AUC 的值似乎不一致？

完整的 output 數據集如下：

真實標簽：

 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0.]

閾值預測

 0. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 0. 0. 0.
 0. 1. 0. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 0. 0. 0. 1. 0. 1.
 1. 1. 0. 0. 0. 0. 1. 1. 1. 1. 1. 1. 1. 0. 0. 0. 0. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 0. 0. 0. 0. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 1. 0. 1. 1. 0. 0. 1.
 1. 1. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0.]

原始乙狀結腸激活 output

 0.80731616 0.81613746 0.63055099 0.33343941 0.33650158 0.26123103
 0.43023067 0.75951926 0.85506684 0.83753035 0.77401593 0.93755849
 0.93669454 0.78037769 0.48196705 0.51107402 0.39020711 0.27603346
 0.27125724 0.79841139 0.96470754 0.97575928 0.9520636  0.98686909
 0.99421722 0.9814615  0.72548573 0.70952273 0.58558095 0.75391273
 0.98747451 0.99592133 0.99348673 0.99636301 0.99966048 0.99927722
 0.93512388 0.87108612 0.76195734 0.45464458 0.44979708 0.3798077
 0.46179509 0.51260215 0.42887223 0.77441987 0.99320274 0.99899955
 0.99885804 0.99888995 0.99996059 0.99992547 0.9893837  0.94771828
 0.90216806 0.63214702 0.70693445 0.62402257 0.72597019 0.72850208
 0.48136757 0.34587109 0.48912585 0.53809234 0.49571105 0.52119752
 0.66452994 0.65721321 0.46201256 0.32531447 0.33560987 0.34733458
 0.54707416 0.66652035 0.67211284 0.64667205 0.77259018 0.81139687
 0.72141833 0.47555719 0.41060125 0.40072988 0.30013099 0.81335717
 0.87926414 0.83410184 0.89994201 0.96761651 0.94806845 0.67343196
 0.60651364 0.57781878 0.76253183 0.95988439 0.98643017 0.98208946
 0.99291688 0.99853936 0.99570023 0.84561008 0.82329192 0.70751861
 0.40768749 0.38326785 0.42332725 0.41978272 0.95580183 0.99577685
 0.99589898 0.99182735 0.99963567 0.99949705 0.98161394 0.93502385
 0.89946262 0.69163107 0.23587978 0.24273368 0.27152508 0.27938265
 0.25957949 0.28954122 0.30340485 0.28367177 0.25412464 0.24931795
 0.40110995 0.38143945 0.49271891 0.50662051 0.33616859 0.52061933
 0.47093576 0.63511254 0.68877464 0.47989569 0.37947267 0.69217007
 0.69413745 0.85119693 0.83831514 0.46003498 0.19595725 0.18322578
 0.13161417 0.17004058 0.155272   0.1832541  0.13801674 0.17109324
 0.16617284 0.16502231 0.16629275 0.17945219 0.18769069 0.19091081
 0.19954858 0.17923033 0.18590597 0.17878488 0.19183244 0.15146982
 0.16887138 0.17444615 0.18757529 0.15070279 0.19910241 0.15885526
 0.18926985 0.19083846 0.1563857  0.19467271 0.19159289 0.21147205
 0.12797629 0.17709421 0.19563617 0.1951601  0.12606692 0.20411101
 0.17489395 0.179219   0.17770813 0.13888956 0.17316737 0.18813291
 0.20011829 0.18280909 0.12445015 0.17259067 0.20987834 0.17725589
 0.18583644 0.16768099 0.17385706 0.19005385 0.16527923 0.17264359
 0.13370521 0.17153564 0.15309515 0.19745554 0.17381944 0.16110312
 0.19662598 0.15733718 0.19763281 0.20617132 0.19089484 0.19732752
 0.1870988  0.16508744 0.13579399 0.13825028 0.19650695 0.2028151
 0.20796896 0.16130049 0.18487175 0.15657099 0.14414533 0.19415208
 0.14158873 0.20252466 0.19986491 0.1761861  0.12490113 0.14082219
 0.19325744 0.17937965 0.17161699 0.20017089 0.1953598  0.19116857
 0.18963095 0.18015937 0.17033672 0.12995853 0.17816802 0.20537938
 0.17656901 0.17246887 0.19970285 0.18360697 0.14851416 0.14957287
 0.17847791 0.19361662 0.12858931 0.15501569 0.16153916 0.18401976
 0.19767486 0.18276181 0.18216812 0.18459979 0.17810379 0.20029616
 0.16008779 0.18842728 0.19535601 0.16842141 0.18356466 0.19130296
 0.19826594 0.16606207 0.17985446 0.18720729 0.16947971 0.19309211
 0.17904012 0.18225684 0.12697826 0.20334946 0.20230229 0.19601187
 0.18372611 0.13250111 0.1508019  0.1991842  0.16360692 0.18059866
 0.17001721 0.16149873 0.16174695 0.19311724 0.17267033 0.14393295
 0.19088417 0.18659356]

Answer 1

這可能是正確的。 准確性取決於閾值，但 auc 不。 Auc 是類分離的力量。 你的 auc = 1 意味着有一個閾值，類可以完美分離，但顯然這個閾值不是 0,5。