工程構(gòu)建

• 導(dǎo)入sklearn相關(guān)包

import numpy as np
from sklearn import neighbors

• 定義readDataSet(fileName, isTest)函數(shù),用于加載訓(xùn)練數(shù)據(jù),返回特征數(shù)據(jù)(dataSet)和標簽(label)

# fileName: 文件名字
# isTest: =True為測試數(shù)據(jù)時,只有54列,label直接設(shè)0
def readDataSet(fileName, isTest):
    fr = open(fileName,encoding='utf-8')
    lines = fr.readlines()

    numLabels = len(lines) 
    labels = np.zeros(numLabels) #用于存放標簽數(shù)據(jù)
    dateSet = np.zeros([numLabels,54],int) #用于存放特征數(shù)據(jù)

    # 逐行讀取數(shù)據(jù)到dataSet,labels
    # 總共55列,前54列是樣本特征,最后一列是樣本類別(label)
    for i in range(numLabels):
        line = lines[i]
        label = 0
        if isTest:  #True:測試集只有54列,最后標簽直接設(shè)為0
            label = 0
        else:  # False:取最后一列為標簽數(shù)據(jù)
            label = line.split(' ')[54]
        labels[i] = label

        dates = np.zeros(54)
        for j in range(53): # 前54列添加為特征數(shù)據(jù)
            dates[j] = line.split(' ')[j]
        dateSet[i] = dates

    fr.close()
    return dateSet,labels

添加isTest參數(shù)是為了區(qū)別訓(xùn)練集(55列)與測試集數(shù)據(jù)(54列)列數(shù)不同,便于交叉驗證時,統(tǒng)計錯誤數(shù)量,核算正確率.

• 加載訓(xùn)練數(shù)據(jù)

train_dataSet,train_labels = readDataSet('data_train.txt', False)

• 構(gòu)建KNN分類器,并調(diào)用fit()函數(shù)

knn = neighbors.KNeighborsClassifier(algorithm='kd_tree', n_neighbors=3)
knn.fit(train_dataSet,train_labels)

• 加載測試集,使用構(gòu)建好的KNN分類器對測試集進行預(yù)測

test_dataSet,test_labels = readDataSet('test.txt', False)
res = knn.predict(test_dataSet)
error = 0
for i in range(len(res)):
    if res[i] != test_labels[i]:
        error += 1
print('error:',error,'正確率:',(len(res) - error) / len(res))

鄰居數(shù)量K影響對比

k = 3時正確率最高,當K > 3 后正確率開始下降