R/Metrics/AUC :이 숫자는 어떻게 AUC가 1이 될 수 있습니까?

Question

나는 다음과 같은 장난감 세트가 : 나는 통계 패키지에서 AUC() 함수를 사용할 때

df1<-structure(list(X1 = c(1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 0, 1, 0, 
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), X2 = c(0.564666517055852, 
0.993248174442609, 0.517237113309667, 0.0128217854167603, 0.952654357126895, 
0.958073009436008, 0.860038905846366, 0.425314512801637, 0.809327038447625, 
0.985049417726494, 0.165982081954436, 0.517237113309667, 0.00211852090504078, 
0.296898500479658, 0.994690775408805, 0.999991149759367, 0.999949243479285, 
0.999979994962211, 0.409697759931823, 0.999995828877373, 0.999991594894354, 
0.999999834424374, 0.952641245900919, 0.999998774453881, 0.999999777896636, 
0.999998864433372, 0.998786297471059, 0.999927421881167, 0.998265361329274, 
0.999550929839182, 0.999900216754163, 0.999912135543067, 0.999999924775596, 
0.996227950775217, 0.998265981873947, 0.999959584436354, 0.999993039255167, 
0.99968139946193, 0.999999997308486, 0.999999458017638, 0.999996417856357, 
0.99958403590535, 0.999998891765696, 0.999999624757926, 0.999818190766803, 
0.999997979863151, 0.999974432439759, 0.996227950775217, 0.999999771762929, 
0.983441425608786, 0.99999843468322)), .Names = c("X1", "X2"), row.names = c(NA, 
-51L), class = "data.frame") 

는이 점수가이 정확하지 않는 것 1.

> Metrics::auc(df1$X1, df1$X2) 
[1] 1 

이라고 하더군요을 . 어떤 제안?

Answer 1

맞습니다. 당신은 (모두 제로는 것보다 낮은 예측 probabillities가) 만 확실히 문제에 가져 가고 3 0을 가지고 있지만, 여기 보면 :

> dat[order(dat[,2]),] 
    X1   X2 
13 0 0.002118521 
4 0 0.012821785 
11 0 0.165982082 
14 1 0.296898500 
19 1 0.409697760 
8 1 0.425314513 
3 1 0.517237113 
12 1 0.517237113 
1 1 0.564666517 
9 1 0.809327038 
7 1 0.860038906 
23 1 0.952641246 
5 1 0.952654357 
6 1 0.958073009 
50 1 0.983441426 
10 1 0.985049418 
2 1 0.993248174 
15 1 0.994690775 
34 1 0.996227951 
48 1 0.996227951 
29 1 0.998265361 
35 1 0.998265982 
27 1 0.998786297 
30 1 0.999550930 
42 1 0.999584036 
38 1 0.999681399 
45 1 0.999818191 
31 1 0.999900217 
32 1 0.999912136 
28 1 0.999927422 
17 1 0.999949243 
36 1 0.999959584 
47 1 0.999974432 
18 1 0.999979995 
16 1 0.999991150 
21 1 0.999991595 
37 1 0.999993039 
20 1 0.999995829 
41 1 0.999996418 
46 1 0.999997980 
51 1 0.999998435 
24 1 0.999998774 
26 1 0.999998864 
43 1 0.999998892 
40 1 0.999999458 
44 1 0.999999625 
49 1 0.999999772 
25 1 0.999999778 
22 1 0.999999834 
33 1 0.999999925 
39 1 0.999999997 

Answer 2

데이터의 ROC를 플로팅 작성자, 당신은 볼 수 있습니다 AUC에가 [ROCR] 같은 다른 패키지 1.0