KNN算法

1.原理版

import numpy as np
import operator

def DataSet():
    group = np.array([[3,104],[2,100],[1,81],[101,10],[99,5],[98,2]])
    labels = ['爱情片','爱情片','爱情片','动作片','动作片','动作片']
    return group,labels

def KNN(in_x,x_labels,y_labels,k):  #in_x是要预测的数据，x_labels是训练数据，y_labels是训练数据的标签，k为k近邻
    x_labels_size = x_labels.shape[0] #获得行数 6
    distances = (np.tile(in_x,(x_labels_size,1)) - x_labels)**2  #np.tile的作用是把in_x弄成6行1列的数组，然后对应位置相加减再平方
    ad_distances = distances.sum(axis=1)  #按列相加求和
    sq_distances = ad_distances ** 0.5    #开方，得到欧式距离
    ed_distances = sq_distances.argsort() #对欧式距离排序，返回索引，就像按学号公开成绩排名
    classdict = {}
    for i in range(k):
        voteI_label = y_labels[ed_distances[i]] #得到排序对应的标签值
        classdict[voteI_label] = classdict.get(voteI_label,0) + 1  #计数，统计前k个里爱情片和动作片的数量
    sort_classdict = sorted(classdict.items(),key=operator.itemgetter(1),reverse=True)   #对classdict.items()进行排序,运用到了operator库
    return sort_classdict[0][0]

if __name__ == '__main__':
    group,labels = DataSet()
    test_x = [18,99]
    print('输入数据所对应的类别是：{}'.format(KNN(test_x,group,labels,3)))

2.代码版

def load_data(folder,domain):
    from scipy import io
    import os
    data = io.loadmat(os.path.join(folder,domain + '_fc6.mat'))
    return data['fts'],data['labels']

def knn_classify(Xs,Ys,Xt,Yt,k=1):
    from sklearn.neighbors import KNeighborsClassifier
    from sklearn.metrics import accuracy_score
    model = KNeighborsClassifier(n_neighbors=k)
    Ys = Ys.ravel()                  #ravel()是 NumPy 数组的方法，用于将多维数组展平为一维数组。
    Yt = Yt.ravel()
    model.fit(Xs,Ys)                 #基于训练数据Xs,Ys拟合model
    Yt_pred = model.predict(Xt)      #用训练好的model对测试数据 Xt预测
    acc = accuracy_score(Yt,Yt_pred)  #accuracy_score 是一个用于计算分类准确率的函数,预测正确的样本数与总样本数之间的比例
    print('Accuracy using kNN :{:.2f}%'.format(acc * 100))

if __name__ == "__main__":
    folder = './dataset/office31-decaf'
    src_domain = 'amazon'     #包括两个表格,一个名称为fts,一个表格名称为labels
    tar_domain = 'webcam'
    Xs,Ys = load_data(folder,src_domain)
    Xt,Yt = load_data(folder,tar_domain)
    print('Source:',src_domain,Xs.shape,Ys.shape)
    print('Target:',tar_domain,Xt.shape,Yt.shape)

    knn_classify(Xs,Ys,Xt,Yt)