机器学习复习：数据处理分析小练习 | 导包

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
%matplotlib inline

一.小练习1

data = {
    'gender':['男','男','male','male','female','女'],
    'age':[17,np.nan,17,16,18,19],
    '语文':['10o',89,32,23,23,23],
    '数学':[23,'-',32,23,23,32],
    '英语':[23,45,32,23,None,32],
#     '体育':[23,45,32,23,Nan,32]
}
df = pd.DataFrame(data = data)
df

	gender	age	语文	数学	英语
0	男	17.0	10o	23	23.0
1	男	NaN	89	-	45.0
2	male	17.0	32	32	32.0
3	male	16.0	23	23	23.0
4	female	18.0	23	23	NaN
5	女	19.0	23	32	32.0

1.量化gender

# 方法一：
df.gender = df.gender.map(lambda x:1 if x =='男'or x=='male' else 0)
df.gender

0    1
1    1
2    1
3    1
4    0
5    0
Name: gender, dtype: int64

# 方法二：
df.gender = df.gender.apply(lambda x:1 if x =='男'or x=='male' else 0)
df.gender

# 方法三：applymap 针对DataFrame 需要花式索引
df['gender'] = df[['gender']].applymap(lambda x:1 if x =='男'or x=='male' else 0)
df

# 方法四：注意小括号是必须的
df.gender = ((df['gender']=='male') | (df['gender']=='男'))*1
df.gender

2.填充空值

平均值填充年龄

df.age.fillna(df['age'].mean(),inplace=True)

成绩的异常值填充为 0

df['英语'].fillna(0,inplace=True)

3.将成绩中的异常值也替换为0

def replace_abnormal(item):
    if type(item)==str:
        # 判断字符串是否全部由数字构成，且在[0,100]之间
        if item.isdigit():
            if(0 <= int(item) <=100):
                return int(item)
            else:
                return 0
        else:
            return 0
    else:
        return item
    

df.iloc[:,2:] = df.iloc[:,2:].applymap(replace_abnormal)
df

	gender	age	语文	数学	英语
0	1	17.0	0	23	23.0
1	1	17.4	89	0	45.0
2	1	17.0	32	32	32.0
3	1	16.0	23	23	23.0
4	0	18.0	23	23	0.0
5	0	19.0	23	32	32.0

4.replace 练习

df.gender.replace({1:'男',0:'女'})

0    男
1    男
2    男
3    男
4    女
5    女
Name: gender, dtype: object

二.小练习2

1.导入数据

data1 = pd.read_csv('./day11复习/data1.txt',
                   na_values='null'
                   )
data1

	Cust_id	x1	x2	Max_ovd_days
0	1	28	36.0	10
1	2	28	36.0	0
2	1	28	36.0	10
3	2	28	36.0	0
4	3	84	8.0	40
5	4	35	8.0	15
6	5	25	NaN	0
7	6	48	15.0	5

data1.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 8 entries, 0 to 7
Data columns (total 4 columns):
 #   Column        Non-Null Count  Dtype  
---  ------        --------------  -----  
 0   Cust_id       8 non-null      int64  
 1   x1            8 non-null      int64  
 2   x2            7 non-null      float64
 3   Max_ovd_days  8 non-null      int64  
dtypes: float64(1), int64(3)
memory usage: 384.0 bytes

2.将x1,x2 中的缺失值替换为平均值

data1.x1.fillna(np.mean(data1.x1),inplace = True)
data1.x2.fillna(np.mean(data1.x2),inplace = True)
data1

	Cust_id	x1	x2	Max_ovd_days
0	1	28	36.0	10
1	2	28	36.0	0
2	1	28	36.0	10
3	2	28	36.0	0
4	3	84	8.0	40
5	4	35	8.0	15
6	5	25	25.0	0
7	6	48	15.0	5

3.生成y 逾期>=30 ->1 ; 其他 -> 0

def func(item):
    if item>=30:
        return 1
    else:
        return 0

data1['y'] = data1.Max_ovd_days.apply(func)
data1

	Cust_id	x1	x2	Max_ovd_days	y
0	1	28	36.0	10	0
1	2	28	36.0	0	0
2	1	28	36.0	10	0
3	2	28	36.0	0	0
4	3	84	8.0	40	1
5	4	35	8.0	15	0
6	5	25	25.0	0	0
7	6	48	15.0	5	0

4.划分数据集

from sklearn.model_selection import train_test_split

X,y = data1.iloc[:,:-1],data1.y
X_train,X_test,y_train,y_test = train_test_split(data1.iloc[:,1:-1],data1.y)

5.网格搜索获得最优参数建模

from sklearn.neighbors import KNeighborsClassifier
from sklearn.model_selection import GridSearchCV
import warnings
warnings.filterwarnings('ignore')

param_grid = {'n_neighbors':[1,3,5,7,9,11,13,15]}
gscv = GridSearchCV(KNeighborsClassifier(),param_grid = param_grid,cv=3)
gscv.fit(X_train,y_train)
print(gscv.best_params_)
# 重新建模
knn = KNeighborsClassifier(n_neighbors=1)
from sklearn.model_selection import cross_val_score
scores = cross_val_score(knn,X,y,cv=3)
scores.mean()

{'n_neighbors': 1}





0.8888888888888888

三:欠采样

# 定义欠采样函数
def RandomUnderSample(x,y,seed,multiple1):
    """
    x,y - 需要欠采样的数据集，必须是DataFrame
    label - 列名
    seed - 种子
    multiple1 - 比例一般是1:1 如果换成1:5 填入5
    """
    
    np.random.seed(seed) # 生成0-1的随机数
    label = y.columns[0] # y.columns:Index(['passed'], dtype='object')
    # 标签值是0或1 
    number0 = len(y[y[label]==0]) # 有多少个等于0
    number1 = len(y[y[label]==1]) # 有多少个标签1
    # 记录数据量少的标签的index值，和数量
    if number0 > number1:
#         min_array1 = np.array(y[y[label]==1].index) # 相应标签值对应的index放到array里
        min_number = number1
    else:
#         min_array1 = np.array(y[y[label]==0].index)
        min_number = number0
    # 另一个部分的label 摘出 和上面数量相同 或者一定比例
    indices_1 = np.array(y[y[label]==1].index)
    indices_0 = np.array(y[y[label]==0].index)
    if len(indices_1) > len(indices_0):
        max_array = indices_1
        min_array = indices_0
    else:
        max_array = indices_0
        min_array = indices_1
    # 从数量多的 随机选择 需要采样的数据
    random_1_indices= np.array(np.random.choice(max_array,min_number*multiple1))
    """
    print "choice([1, 2, 3, 5, 9]) : ", random.choice([1, 2, 3, 5, 9])
    """
    # 
    index = np.concatenate([min_array1,random_1_indices])
    #
    X_under_sample = x.loc[index,:]
    Y_under_sample = y.loc[index,:]
    
    return X_under_sample,Y_under_sample

四.导包

基础导入

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

plt.rcParams['font.sans-serif'] = 'Songti SC'
plt.rcParams['axes.unicode_minus'] = False

%config InlineBackend.figure_format = 'svg'

import warnings
warnings.filterwarnings('ignore')

机器学习方法总结

线性回归，线性分类

from sklearn.linear_model import LinearRegression,Lasso,Ridge

from sklearn.linear_model import LogisticRegression,SGDClassifier

KNN

from sklearn.neighbors import KNeighborsClassifier,KNeighborsRegressor

KMeans

from sklearn.cluster import KMeans

贝叶斯

from sklearn.naive_bayes import GaussianNB,MultinomialNB,BernoulliNB

决策树

from sklearn.tree import DecisionTreeClassifier,DecisionTreeRegressor

支持向量机

from sklearn.svm import SVR,SVC 

集成学习方法

from sklearn.ensemble import BaggingClassifier,BaggingRegressor

from sklearn.ensemble import RandomForestClassifier,RandomForestRegressor

from sklearn.ensemble import ExtraTreesClassifier,ExtraTreesRegressor

序列学习方法

from sklearn.ensemble import AdaBoostRegressor,AdaBoostClassifier

from sklearn.ensemble import GradientBoostingRegressor,GradientBoostingClassifier

from xgboost import XGBRegressor,XGBClassifier

特征选择

Filter 基于方差选择

from sklearn.feature_selection import VarianceThreshold

Wrapper

from sklearn.feature_selection import RFE

Embeded

基于惩罚项的特征选择法

from sklearn.feature_selection import SelectFromModel

基于树模型的特征选择法

from sklearn.feature_selection import SelectFromModel
from sklearn.ensemble import GradientBoostingClassifier
# GBDT作为基模型的特征选择
# SelectFromModel(GradientBoostingClassifier()).fit_transform(iris.data, iris.target)

区间缩放

from sklearn.preprocessing import MinMaxScaler

标准化

from sklearn.preprocessing import StandardScaler

归一化

from sklearn.preprocessing import Normalizer

对定量特征二值化

from sklearn.preprocessing import Binarizer

对定性特征哑编码

from sklearn.preprocessing import OneHotEncoder
from sklearn.preprocessing import LabelEncoder

PCA

from sklearn.decomposition import PCA

LDA

from sklearn.discriminant_analysis import LinearDiscriminantAnalysis

网格搜索

from sklearn.model_selection import GridSearchCV

交叉验证

from sklearn.model_selection import cross_val_score

集成学习

from sklearn.ensemble import VotingClassifier

回归器性能评估

from sklearn.metrics import mean_squared_error
from sklearn.metrics import mean_absolute_error
from sklearn.metrics import explained_variance_score
from sklearn.metrics import r2_score

`