sklearn KFold()

大家好，又见面了，我是你们的朋友全栈君。

最近实践过程中遇到需要KFold()
记录一下，以便日后查阅
KFold()在sklearn中属于model_slection模块

from sklearn.model_selection import KFold

KFold(n_splits=’warn’, shuffle=False, random_state=None)
参数:
n_splits 表示划分为几块（至少是2）

shuffle 表示是否打乱划分，默认False，即不打乱

random_state 表示是否固定随机起点，Used when shuffle == True.

方法
1，get_n_splits([X, y, groups]) 返回分的块数
2，split(X[,Y,groups]) 返回分类后数据集的index

例子：
1, get_n_splits()

from sklearn.model_selection import KFold
X = np.array([[1, 2], [3, 4], [1, 2], [3, 4]])
y = np.array([1, 2, 3, 4])
kf = KFold(n_splits=2)
print( kf.get_n_splits(X))

输出2

2, split()

for train_index, test_index in kf.split(X):
	print("TRAIN:", train_index, "TEST:", test_index)
	X_train, X_test = X.iloc[train_index], X.iloc[test_index]
	y_train, y_test = y.iloc[train_index], y.iloc[test_index]

TRAIN: [2 3] TEST: [0 1]
TRAIN: [0 1] TEST: [2 3]

for k,(train,test) in enumerate(kf.split(x,y)):
	print (k,(train,test))
	x_train=X.iloc[train]
	x_test=X.iloc[test]
	y_train=Y.iloc[train]
	y_test=Y.iloc[tes]

用KFold()优化逻辑回归参数C
参数C为正则化项的系数gama的倒数（C=1/gama）

def best_C_param (x,y):
    kf=KFold(n_splits=7,shuffle=True,random_state=0)
    C_param=[0.001,0.01,0.1,1,10,100]
    result=[]
    for c_parm in C_param:
        print('C_param',c_parm)
        recall_score_lr_kf=[]
        for k,(train,test) in enumerate(kf.split(x,y)):
            #print(y.iloc[train])
            model = LogisticRegression(C=c_parm,penalty='l2')
            lr_kf=model.fit(x.iloc[train],y.iloc[train].values.ravel())
            pred_lr_kf=lr_kf.predict(x.iloc[test])
            recall_score_lrkf=recall_score(y.iloc[test],pred_lr_kf)
            recall_score_lr_kf.append(recall_score_lrkf)
            print('iteration',k,'recall score',recall_score_lrkf)
        result.append(np.mean(recall_score_lr_kf))
        print(c_parm,np.mean(recall_score_lr_kf))
        print('bets mean recall score',max(result))


print('using whole datasets X and Y')
best_C_param(X,Y)

print('-------------------------')
print('using under sampling data X and Y')
best_C_param(X_underSampling,Y_underSampling)

一点小小的感触：
数据是非平衡数据结构，正样本1在总体数据集中只占有0.17%
欠采样处理后，二分类比例达到1：1
欠采样处理后的数据KFold寻找LR的最佳C：

方法1：不打乱划分，即shuffle=False (默认)，其他同上

kf=KFold(n_splits=7）

每个C参数，出现recall score为0 （7块中至少2块为0），导致平均下来每个c_parm的recall均分只有0.5左右。
原因：不打乱的时候，分块中有些没分到正样本

方法2：打乱划分，固定随机种子

kf=KFold(n_splits=7,shuffle=True,random_state=0)

输出：结果对欠采样处理后的数据表现较好

using whole datasets X and Y before under sampling

C_param 0.001

iteration 0 recall score 0.5797101449275363

iteration 1 recall score 0.5223880597014925

iteration 2 recall score 0.5

iteration 3 recall score 0.5571428571428572

iteration 4 recall score 0.6774193548387096

iteration 5 recall score 0.5692307692307692

iteration 6 recall score 0.4567901234567901

0.001 0.5518116156140221

bets mean recall score 0.5518116156140221

C_param 0.01

iteration 0 recall score 0.5942028985507246

iteration 1 recall score 0.5671641791044776

iteration 2 recall score 0.5512820512820513

iteration 3 recall score 0.6142857142857143

iteration 4 recall score 0.6774193548387096

iteration 5 recall score 0.6

iteration 6 recall score 0.5185185185185185

0.01 0.5889818166543136

bets mean recall score 0.5889818166543136

C_param 0.1

iteration 0 recall score 0.6231884057971014

iteration 1 recall score 0.6119402985074627

iteration 2 recall score 0.5512820512820513

iteration 3 recall score 0.6142857142857143

iteration 4 recall score 0.6935483870967742

iteration 5 recall score 0.6153846153846154

iteration 6 recall score 0.5555555555555556

0.1 0.6093121468441821

bets mean recall score 0.6093121468441821

C_param 1

iteration 0 recall score 0.6521739130434783

iteration 1 recall score 0.6119402985074627

iteration 2 recall score 0.5512820512820513

iteration 3 recall score 0.6428571428571429

iteration 4 recall score 0.7096774193548387

iteration 5 recall score 0.6153846153846154

iteration 6 recall score 0.5802469135802469

1 0.6233660505728338

bets mean recall score 0.6233660505728338

C_param 10

iteration 0 recall score 0.6521739130434783

iteration 1 recall score 0.6119402985074627

iteration 2 recall score 0.5512820512820513

iteration 3 recall score 0.6428571428571429

iteration 4 recall score 0.7096774193548387

iteration 5 recall score 0.6153846153846154

iteration 6 recall score 0.5802469135802469

10 0.6233660505728338

bets mean recall score 0.6233660505728338

C_param 100

iteration 0 recall score 0.6521739130434783

iteration 1 recall score 0.6119402985074627

iteration 2 recall score 0.5512820512820513

iteration 3 recall score 0.6428571428571429

iteration 4 recall score 0.7096774193548387

iteration 5 recall score 0.6153846153846154

iteration 6 recall score 0.5802469135802469

100 0.6233660505728338

bets mean recall score 0.6233660505728338

 

-------------------------

using under sampling data X and Y

C_param 0.001

iteration 0 recall score 0.9558823529411765

iteration 1 recall score 0.9861111111111112

iteration 2 recall score 0.9473684210526315

iteration 3 recall score 0.9466666666666667

iteration 4 recall score 0.9661016949152542

iteration 5 recall score 0.96

iteration 6 recall score 0.9701492537313433

0.001 0.9617542143454548

bets mean recall score 0.9617542143454548

C_param 0.01

iteration 0 recall score 0.9558823529411765

iteration 1 recall score 0.8888888888888888

iteration 2 recall score 0.881578947368421

iteration 3 recall score 0.8666666666666667

iteration 4 recall score 0.9661016949152542

iteration 5 recall score 0.9466666666666667

iteration 6 recall score 0.9253731343283582

0.01 0.9187369073964903

bets mean recall score 0.9617542143454548

C_param 0.1

iteration 0 recall score 0.9264705882352942

iteration 1 recall score 0.8888888888888888

iteration 2 recall score 0.868421052631579

iteration 3 recall score 0.8666666666666667

iteration 4 recall score 0.9661016949152542

iteration 5 recall score 0.9466666666666667

iteration 6 recall score 0.8955223880597015

0.1 0.9083911351520072

bets mean recall score 0.9617542143454548

C_param 1

iteration 0 recall score 0.9264705882352942

iteration 1 recall score 0.9027777777777778

iteration 2 recall score 0.881578947368421

iteration 3 recall score 0.8933333333333333

iteration 4 recall score 0.9491525423728814

iteration 5 recall score 0.9466666666666667

iteration 6 recall score 0.8955223880597015

1 0.9136431776877252

bets mean recall score 0.9617542143454548

C_param 10

iteration 0 recall score 0.9264705882352942

iteration 1 recall score 0.8888888888888888

iteration 2 recall score 0.8947368421052632

iteration 3 recall score 0.9066666666666666

iteration 4 recall score 0.9491525423728814

iteration 5 recall score 0.9466666666666667

iteration 6 recall score 0.8955223880597015

10 0.9154435118564803

bets mean recall score 0.9617542143454548

C_param 100

iteration 0 recall score 0.9264705882352942

iteration 1 recall score 0.9027777777777778

iteration 2 recall score 0.881578947368421

iteration 3 recall score 0.9066666666666666

iteration 4 recall score 0.9661016949152542

iteration 5 recall score 0.96

iteration 6 recall score 0.8805970149253731

100 0.9177418128412552

bets mean recall score 0.9617542143454548

 

Process finished with exit code 0