机器学习之KNN学习曲线

实例判断约会匹配精准度

1、并未进行标准化或者归一化

import pandas as pd
import warnings
warnings.simplefilter("ignore")
from sklearn.neighbors import KNeighborsClassifier
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import MinMaxScaler,StandardScaler  # 引入归一化与标准化
df = pd.read_csv('C:/*************************/datingTestSet.txt',header=None,sep='\t')
print(df.head())
# 获取特征
feature = df[[0, 1, 2]]
# 获取标签
target = df[3]
print(feature.head())
print(target.head())
# 在进行提取特征之前要进行查看他则会那个是否是二维的
print(feature.shape)
x_train,x_test,y_train,y_test = train_test_split(feature,target,train_size=0.666,random_state=2233)
# 训练模型
knn = KNeighborsClassifier(n_neighbors=18)
# 传入要训练的模型的特征与标签
knn.fit(x_train,y_train)
# 查看训练精度
scope = knn.score(x_test,y_test)
print(scope)

2、进行归一化

需要在开头导入包
from sklearn.preprocessing import MinMaxScaler
以下需要改的代码就是，将特征进行归一化处理，标签不用

mm = MinMaxScaler()
m_feature = mm.fit_transform(feature)
knn2 = KNeighborsClassifier(n_neighbors=18)
x2_train,x2_test,y2_train,y2_test = train_test_split(m_feature,target,train_size=0.8,random_state=2525)
knn2.fit(x2_train,y2_train)
scope2 = knn2.score(x2_test,y2_test)
print(scope2)

得到的结果：

       0          1         2           3
0  40920   8.326976  0.953952  largeDoses
1  14488   7.153469  1.673904  smallDoses
2  26052   1.441871  0.805124   didntLike
3  75136  13.147394  0.428964   didntLike
4  38344   1.669788  0.134296   didntLike
       0          1         2
0  40920   8.326976  0.953952
1  14488   7.153469  1.673904
2  26052   1.441871  0.805124
3  75136  13.147394  0.428964
4  38344   1.669788  0.134296
0    largeDoses
1    smallDoses
2     didntLike
3     didntLike
4     didntLike
Name: 3, dtype: object
(1000, 3)
0.8083832335329342
0.92
Process finished with exit code 0

3、进行标准化

需要导入的模块是
from sklearn.preprocessing import StandardScaler
只需要更改特征即可

import pandas as pd
import warnings
warnings.simplefilter("ignore")
from sklearn.neighbors import KNeighborsClassifier
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import MinMaxScaler,StandardScaler  # 引入归一化与标准化
df = pd.read_csv('C:/Users/lenovo/Desktop/python_use/data/datasets/datingTestSet.txt',header=None,sep='\t')
print(df.head())
# 获取特征
feature = df[[0, 1, 2]]
# 获取标签
target = df[3]
print(feature.head())
print(target.head())
# 在进行提取特征之前要进行查看他则会那个是否是二维的
print(feature.shape)
x_train,x_test,y_train,y_test = train_test_split(feature,target,train_size=0.666,random_state=2233)
# 训练模型
knn = KNeighborsClassifier(n_neighbors=18)
# 传入要训练的模型的特征与标签
knn.fit(x_train,y_train)
# 查看训练精度
scope = knn.score(x_test,y_test)
print(scope)
# 归一化
mm = MinMaxScaler()
m_feature = mm.fit_transform(feature)
knn2 = KNeighborsClassifier(n_neighbors=18)
x2_train,x2_test,y2_train,y2_test = train_test_split(m_feature,target,train_size=0.8,random_state=2525)
knn2.fit(x2_train,y2_train)
scope2 = knn2.score(x2_test,y2_test)
print(scope2)
# 标准化
std = StandardScaler()
std_feature = std.fit_transform(feature)
knn3 = KNeighborsClassifier(n_neighbors=18)
x3_train,x3_test,y3_train,y3_test=train_test_split(std_feature,target)
knn3.fit(x3_train,y3_train)
scope3 = knn3.score(x3_test,y3_test)
print(scope3)

得到的结果

0.8083832335329342
0.92
0.948

可以发现
标准化后的准确率更高，
以后我在选择实验是不是也可以这样

模型的超参数：
如果说模型里面的参数的变化影响模型的精度，我们就称为模型的超参数
但是我们也不能逐次的调整参数，这个时候就要想办法找出最好的参数最好不用试了，根据学习曲线找到最合适的参数
需要先导入画图包

import matplotlib.pyplot as plt
ks = np.arange(1, 100, 5)#确定要试的参数范围，并进行取值
print(ks)
scores = []#找到一个数组接收测试结果
for k in ks:
    knn4 = KNeighborsClassifier(k)
    knn4.fit(x3_train,y3_train)
    score = knn4.score(x3_test,y3_test)
    scores.append(score)
plt.plot(ks,scores)
plt.xlabel('k')
plt.ylabel('s')
plt.show()

输出结果：

(1000, 3)
0.8083832335329342
0.92
0.932
[ 1  6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96]

在图里面找到最合适的点就好
scores = np.array(scores)# 将scores转换成np数组
先找到scores的最大值的索引
然后将索引放入ks中就可以的到训练效果最好的参数
max_value_index =(scores.argmax())
print(ks[max_value_index])