相关日志
-
-
分享
每日学习
-
lww1993 2014-4-8 11:04
-
1.ROC曲线:(receiver operating characteristic curve): 纵轴:true positive/positive 横轴:false positive/negative 2.Precision and Recall: Precision=true positive/claimed positive Recall= true positive/all positive Here are 9 dogs and 7 cats. You want to select the dogs. If you select 10, 6 of which are dogs. Then the precision=6/10 while the recall is 6/9. 3.stratified sampling strategy. 一个很好地处理数据不平衡的方法; 4.(4.29)natural smooth spline: A.所有的点都视为knots,两点之间是用三次多项式去拟合的; B.在曲线的中间的knots处都是二次连续可微的; C.在曲线两边的knots处,只要求线性; D: 上式中,f是我们期望的估计函数,x,y是相应的点,lambda是系数,第一项是看估计的好坏,最后一项是惩罚曲线的弯曲程度,这个式子的最佳估计正是一个natural cubic spline. 此时我们可以将其基代入,计算出相应的系数。详见“Elements of statistical learning”,second edition,5.1到5.4
-
个人分类: R语言学习|31 次阅读|0 个评论
GMT+8, 2025-12-24 18:52