本文通过海伦约会的例子来测试之前写的KDTree的效果，并且探讨了特征是否进行归一化对整个模型的表现的影响。
最后发现在机器学习中，特征归一化确实对模型能提供非常大的帮助。
1 from KDTree import KDTree # 参考实现KDtree的随笔
2 from sklearn import model_selection,preprocessing
3 import pandas as pd
4 class KNN(object):
5 def __init__(self,K=1,p=2):
6 self.kdtree= KDTree()
7 self.K =K
8 self.p=p
9 def fit(self,x_data,y_data):
10 self.kdtree.build_tree(x_data,y_data)
11 def predict(self,pre_x,label):
12 if 'class' in label:
13 return self.kdtree.predict_classification(pre_x,K=self.K)
14 else :
15 return self.kdtree.predict_regression(pre_x,K=self.K)
16 def test_check(self,test_xx,test_y):
17 # only support classification problem
18 correct =0
19 for i,xi in enumerate(test_xx):
20 pre_y = self.kdtree.predict_classification(Xi=xi,K=self.K)
21 if pre_y == test_y[i]:
22 correct+=1
23 return correct/len(test_y)
24
25
26 file_path = "datingTestSet.txt"
27 data = pd.read_csv(file_path, sep="\t",header=None)
28 XX = data.iloc[:,:-1].values
29 Y = data.iloc[:,-1].values
30 train_xx , test_xx, train_y,test_y = model_selection.train_test_split(XX,Y,test_size= 0.2,random_state=123,shuffle=True)
31 knn=KNN(K=5,p=2)
32 knn.fit(train_xx,train_y)
33 acc = knn.test_check(test_xx,test_y)
34 print("No Standard Scale Accuracy: ",acc)
35 # 考虑到数据中不同维度之间的数值相差过大，进行特征缩放
36 scaler = preprocessing.StandardScaler()
37 # 计算均值和标准差只能用训练集的数据
38 scaler.fit(train_xx)
39 stand_train_xx = scaler.transform(train_xx)
40 stand_test_xx = scaler.transform(test_xx)
41 new_knn = KNN(K=5,p=2)
42 new_knn.fit(stand_train_xx,train_y)
43 new_acc = new_knn.test_check(stand_test_xx,test_y)
44 print("Standard Scale Accuracy: ",new_acc)
