Created by: JingChunzhen

使用如下 api 进行 inference model 计算 auc，batch_size == 100

# paddle 1.5.2
global_auc, batch_auc, _ = fluid.layers.auc(input=pred, label=input_label, slide_steps=1)

loss batch_auc global_auc

0 1.86869418621 [0.] [0.] 1 1.95746719837 [0.97979798] [0.98492462] # ？ 2 1.95067119598 [0.] [0.98327759] 3 1.92995178699 [0.] [0.98746867] 4 1.94889438152 [0.] [0.98797595] 5 1.85230088234 [1.] [0.99190396] 6 2.05646729469 [0.98989899] [0.99155891] 7 1.89979791641 [0.98989899] [0.99144654] 8 1.82984125614 [0.] [0.9922905] 9 1.674020648 [0.98989899] [0.99010731] 10 1.9083533287 [0.] [0.9900975] 11 1.83100903034 [0.98989899] [0.99024069] 12 1.82303655148 [0.] [0.9907745] 13 1.82914912701 [0.95959596] [0.98715876] 14 1.83058965206 [1.] [0.98912752] 15 1.96960937977 [0.] [0.98902516] 16 1.80508208275 [0.] [0.98931953]

问题： 部分 batch_auc == 0 , 但是 global auc 维持 在高位 ？