（五）K近邻算法(KNN)

一、鸢尾花案例

import numpy as np
import matplotlib.pylab as plt
from sklearn import datasets

iris = datasets.load_iris()
iris.keys()

dict_keys(['data', 'target', 'target_names', 'DESCR', 'feature_names', 'filename']) 

x = iris.data
y = iris.target

print("x",x.shape)
print("y",y.shape)

x (150, 4)
y (150,) 

 1.1 训练集和测试集拆分

from sklearn.model_selection  import train_test_split

# test_size：测试集占比
# random_state：随机种子
X_train, X_test, y_train, y_test = train_test_split(x,y,test_size=0.2,random_state=666)

1.2 未进行数据归一化 

from  sklearn.neighbors  import KNeighborsClassifier

clf1 = KNeighborsClassifier(n_neighbors=3)

clf1.fit(X_train,y_train)

 KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski', metric_params=None, n_jobs=None, n_neighbors=3, p=2, weights='uniform')

clf1.score(X_test,y_test)   # 预期得分为1

1.3 数据归一化：StandardScaler

• 为什么数据归一化：将所有数据映射到同一尺度

•  最值归一化：所有数据映射到0-1之间。适用于分布有明显边界的情况，受离群值影响较大

• 均值方差归一化：所有数据归一到均值为0、方差为1的分布中。适用于分布没有明显边界，可能存在极端值的情况

from sklearn.preprocessing import StandardScaler

# 均值方差归一化
scaler = StandardScaler()

# 使用X_train得到均值方差归一化参数，比如std、mean等
scaler.fit(X_train)

 StandardScaler(copy=True, with_mean=True, with_std=True)

# 均值
scaler.mean_     # array([5.83416667, 3.08666667, 3.70833333, 1.17      ])

# 数据分布情况（标准差）
scaler.scale_    # array([0.81019502, 0.44327067, 1.76401924, 0.75317107])

# 对训练数据进行归一化
X_train = scaler.transform(X_train)

# 对测试数据进行归一化
X_test = scaler.transform(X_test)

from  sklearn.neighbors  import KNeighborsClassifier

clf = KNeighborsClassifier(n_neighbors=3)

clf.fit(X_train,y_train)

KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski', metric_params=None, n_jobs=None, n_neighbors=3, p=2, weights='uniform') 

# 准确率计算
clf.score(X_test,y_test)   #  1.0

 二、手写数字图片案例

import numpy as np
import matplotlib.pylab as plt
from sklearn import datasets
from sklearn.model_selection  import train_test_split

digits = datasets.load_digits()

digits.keys()

dict_keys(['data', 'target', 'target_names', 'images', 'DESCR'])

X = digits.data
y = digits.target

print(X.shape)   #   (1797, 64)
print(y.shape)   #   (1797,)

2.1 训练集、测试集拆分

# test_size：测试集占比
# random_state：随机种子
X_train, X_test, y_train, y_test = train_test_split(X,y,test_size=0.2,random_state=666)

from sklearn.neighbors  import KNeighborsClassifier

# 创建分类器
clf = KNeighborsClassifier(n_neighbors=3)

2.2 训练

clf.fit(X_train, y_train)

KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',
                     metric_params=None, n_jobs=None, n_neighbors=3, p=2,
                     weights='uniform') 

2.3 预期

y_predict = clf.predict(X_test)

from sklearn.metrics import accuracy_score

# accuracy 精确度得分
accuracy_score(y_test,y_predict)   #  0.9888888888888889

# 不经过测试集测试，直接算出模型得分
clf.score(X_test,y_test)   #  0.9888888888888889

 2.4 Grid Search 网格搜索，查询最优模型参数

# weights=uniform  所有的点具有相同的权重

# weights=distance时，p才有意义
'''
权重是距离反比，距离较近的点比距离远的点影响更大
p:整数，可选，默认值2
闵科夫斯基距离的指数，
当p=1，等价于使用曼哈顿距离
当p=2，等价于欧式距离 
'''
param = [
    {
        'weights':['uniform'],
        'n_neighbors':[i for i in range (1,11)]
    },
    {
        'weights':['distance'],
        'n_neighbors':[i for i in range (1,11)],
        'p':[i for i in range (1,6)]
    }
]

说明：cross-validation 交叉验证，将训练数据分成多分，分别取出一部分进行训练，另一部分进行调参，最后用测试数据进行验证。这样可以有效避免在调参阶段模型对测试数据过拟合

knn_clf = KNeighborsClassifier()

#  cross-validation
from sklearn.model_selection import GridSearchCV

%%time
# n_jobs：分配多少个核给它，-1表示所有
# verbose：过程输出，数据越大输出的信息越多越详细
search = GridSearchCV(knn_clf, param, n_jobs=-1,verbose=4)
search.fit(X_train, y_train)

# 最优模型参数，返回最优模型
search.best_estimator_

'''
KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',
                     metric_params=None, n_jobs=None, n_neighbors=1, p=2,
                     weights='uniform')
'''

# 该模型最好的准确度
search.best_score_    #  0.9860820751064653

# 最佳参数
search.best_params_   #  {'n_neighbors': 1, 'weights': 'uniform'}

# 最优模型赋值给knn_clf
knn_clf = search.best_estimator_

knn_clf.score(X_test, y_test)  #  0.9833333333333333