ML Metrics CrossValidation

[junxnone]

CrossValidation 交叉验证

Reference

• 几种交叉验证（cross validation）方式的比较
• sklearn - cross validation
• sklearn.model_selection 介绍

Brief

• Cross Validation - 交叉验证

Num	Cross Validation	Description
1	K Fold
2	Repeated K-Fold
3	Leave One Out (LOO)
4	Leave P Out (LPO)
5	Shuffle & Split
6	Stratified @12345
7	Group @12345

Standard Cross Validation - K Fold Cross Validation

• data split K part
• x ∈ [1, K] part 作为 Test, 剩下的 K-1 part 作为 training

  K = 5 / 10

Stratified k-fold cross validation - 分层交叉验证

• 每一折中都保持着原始数据中各个类别的比例关系

Leave-one-out Cross-validation 留一法

• 样本容量为n，则k=n，进行n折交叉验证，每次留下一个样本进行验证
• 主要针对小样本数据

Shuffle-split cross-validation

[junxnone]

junxnone/ml#6