O'Reilly logo
  • tim hsu thinks this is interesting:

You may wonder why the precision curve is bumpier than the recall curve


Cover of Hands-On Machine Learning with Scikit-Learn and TensorFlow


Precision/ Recall Curve. 裡面。 threshold 上升, precision 不一定會一直提升。 請先記住,precision 的公式是 tp/(tp+fp) threshold 只上升一點點時,可能會減少一個 tp. 但是不對 fp 有任何影響。 此時, precision 就會下降一點。