KFold與StratifiedKFold的區別

class sklearn.model_selection.StratifiedKFold(n_splits=3, shuffle=False, random_state=None)
Stratified K-Folds cross-validator Provides train/test indices to split data in train/test sets.This cross-validation object is a variation of KFold that returns stratified folds. The folds are made by preserving the percentage of samples for each class

意思就是Stra~是按著不同類別標籤的相對佔比來進行的分折

看看程式碼

import numpy as np
import sklearn.datasets as ds
import sklearn.svm as sksvm
import sklearn.model_selection as skmodsel

a=[[i] for i in range(100)]
b=[0]*30;b.extend([1]*30);b.extend([2]*30);b.extend([3]*10)

K_strafold=skmodsel.StratifiedKFold(n_splits=10) 
K_folds=skmodsel.KFold(n_splits=10
 
)
for train_indices, test_indices in K_folds.split(a):
    d={}
    for i in train_indices:
        d[b[i]]=d.setdefault(b[i],0)+1
    print(d)

print('------------------')    
for train_indices, test_indices in K_strafold.split(a,b):
    d={}
    for i in train_indices:
        d[b[i]]=d.setdefault(b[i],0
 
)+1
    print(d)

d={}
for i in range(100):
    d[b[i]]=d.setdefault(b[i],0)+1   
print(d)

結果如下

{0: 20, 1: 30, 2: 30, 3: 10}
{0: 20, 1: 30, 2: 30, 3: 10}
{0: 20, 1: 30, 2: 30, 3: 10}
{0: 30, 1: 20, 2: 30, 3: 10}
{0: 30, 1: 20, 2: 30, 3: 10}
{0: 30, 1: 20, 2: 30, 3: 10}
{0: 30, 1: 30, 2: 20, 3: 10}
{0: 30, 1: 30, 2: 20, 3: 10}
{0: 30, 1: 30, 2: 20, 3: 10}
{0: 30, 1: 30, 2: 30}
------------------
{0: 27, 1: 27, 2: 27, 3: 9}
{0: 27, 1: 27, 2: 27, 3: 9}
{0: 27, 1: 27, 2: 27, 3: 9}
{0: 27, 1: 27, 2: 27, 3: 9}
{0: 27, 1: 27, 2: 27, 3: 9}
{0: 27, 1: 27, 2: 27, 3: 9}
{0: 27, 1: 27, 2: 27, 3: 9}
{0: 27, 1: 27, 2: 27, 3: 9}
{0: 27, 1: 27, 2: 27, 3: 9}
{0: 27, 1: 27, 2: 27, 3: 9}
------------------
{0: 30, 1: 30, 2: 30, 3: 10}

我們設定的是十折，也就是說將a分為十份，每次取一份來做test集
結果很明顯，類別標籤裡有0,1,2各30個，和10個3標籤，他們的佔比就是3:3:3:1
當直接使用KFold的時候，則是每次取出十個同一列表標籤的來做一折，看起來就好像是按著b的排列順序，第一次取0~9做第一折，10-19做第二折，以此類推
但當使用Stra~的時候，每次取都是按著佔比來取的，每一折都滿足3:3:3:1，測試集也是滿足這種