机器学习——测试K近邻算法的性能

我们怎么评价得到的K近邻算法的性能？

K近邻算法训练后直接投入真实环境中使用可能会造成真实损失，我们需要判断该算法的准确度以及性能，怎么测试？

解决方法：将原始的训练数据分隔成两部分，一部分作为测试数据，一部分作为训练数据，即train_test_spilt

代码如下：

import math
import numpy as np
from collections import Counter
import matplotlib.pyplot as plt
from sklearn import datasets
plt.rcParams['font.sans-serif']=['SimHei'] #显示中文标签
plt.rcParams['axes.unicode_minus']=False #对负号进行设置
class KNNClassfiy(object):
    def __init__(self,k):
        assert k>=1,'k must be valid'
        self.k=k
        self._xTrain=None
        self._yTrain=None


    def fit(self,xTrain,yTrain):
        assert xTrain.shape[0]==yTrain.shape[0],
            'The size of xTrain must be equals to the size of yTrain'
        assert self.k>=1 and self.k<=xTrain.shape[0],
            'The size of xTrain must be least at k'
        self._xTrain=xTrain
        self._yTrain=yTrain
        return self

    def predict(self,X_predict):
        # x是预测数据
        assert X_predict.shape[1]==self._xTrain.shape[1],
            'The feature of x must be equal to xTrain'
        assert self._xTrain is not None and self._yTrain is not None,
            'must fit before predict'
        y_predict=[self._predict(x) for  x in X_predict]
        return np.array(y_predict)

    def _predict(self,x):
        distances=[math.sqrt(np.sum((xTrain-x)**2)) for xTrain in self._xTrain]
        nearest=np.argsort(distances)
        top_y=[self._yTrain[i] for i in nearest[:self.k]]
        votes=Counter(top_y)
        return votes.most_common(1)[0][0]
    def __repr__(self):
        return self.k

def train_test_spilt(x,y,rate=0.2):
    assert x.shape[0]==y.shape[0],'x must be same as y in shape[0]'
    test_size=int(len(x)*rate)
    # 测试数据的大小
    shuff_index=np.random.permutation(len(x))
    test_index=shuff_index[:test_size]
    # 前面是测试数据的下标
    train_index=shuff_index[test_size:]
    # 后面是训练数据的下标
    return x[test_index],x[train_index],y[test_index],y[train_index]


iris=datasets.load_iris();
x=iris.data
y=iris.target
x_test,x_train,y_test,y_train=train_test_spilt(x,y,0.2)

KNN_clf=KNNClassfiy(k=6);
xTrain=x_train
yTrain=y_train
KNN_clf.fit(xTrain=xTrain,yTrain=yTrain)
x_predict=np.array(x_test)

# 绘制出来
plt.scatter(xTrain[:,0],xTrain[:,1],label='训练数据',c='r',s=75)
plt.scatter(x_predict[:,0],x_predict[:,1],label='测试数据',c='b',s=75)
plt.legend()
res=KNN_clf.predict(x_predict)
percent=sum(res==y_test)/len(x_test)
print(percent)

没有设置随机数种子，当多次运行代码时，percent不一定总是相等，可以添加seed来改进。

绘制的散点图