本文主要是介绍共享单车线性回归分析，希望对大家解决编程问题提供一定的参考价值，需要的开发者们随着小编来一起学习吧！
给出共享单车的相关业务数据，建立kmeans模型，不同维度分群结果进行分析
| column | definition |
|---|---|
| user_id | 用户编码 |
| start_time | 开始时间 |
| end_time | 结束时间 |
| timeduration | 骑行时长 |
| bikeid | 自行车编码 |
| tripduration | 骑行距离 |
| from_station_id | 开始站编码 |
| from_station_name | 开始站名字 |
| to_station_id | 结束站编码 |
| to_station_name | 结束站名字 |
| usertype | 用户种类 |
| gender | 性别 |
| birthyear | 出生年份 |
| age | 年龄 |
import pandas as pd
mobike=pd.read_csv("mobike.csv")
mobike.head() #查看数据内容
mobike.info() #查看数据类型，观察是否有缺失值<class 'pandas.core.frame.DataFrame'>
RangeIndex: 6427 entries, 0 to 6426
Data columns (total 15 columns):# Column Non-Null Count Dtype
--- ------ -------------- ----- 0 Unnamed: 0 6427 non-null int64 1 user_id 6427 non-null int64 2 start_time 6427 non-null object 3 end_time 6427 non-null object 4 timeduration 6427 non-null int64 5 bikeid 6427 non-null int64 6 tripduration 6427 non-null object 7 from_station_id 6427 non-null int64 8 from_station_name 6427 non-null object 9 to_station_id 6427 non-null int64 10 to_station_name 6427 non-null object 11 usertype 6427 non-null object 12 gender 5938 non-null object 13 birthyear 5956 non-null float6414 age 6427 non-null object
dtypes: float64(1), int64(6), object(8)
memory usage: 753.3+ KBmobike.describe() #结果的索引将包括计数，平均值，标准差，最小值，最大值以及较低的百分位数和50。
单变量分析
#删除无实际意义的索引行，inplace=True是直接在原始数据中保存排序之后的结果
mobike.drop("Unnamed: 0",inplace=True,axis=1)#类别型变量转换成数字型
#mobike['age']，对于age这个列做修改后的存储列
mobike['age']=mobike.age.str.replace('\'','').replace(' ','0').astype(int)
mobike.info()#此时的age变成int32类型<class 'pandas.core.frame.DataFrame'>
RangeIndex: 6427 entries, 0 to 6426
Data columns (total 14 columns):# Column Non-Null Count Dtype
--- ------ -------------- ----- 0 user_id 6427 non-null int64 1 start_time 6427 non-null object 2 end_time 6427 non-null object 3 timeduration 6427 non-null int64 4 bikeid 6427 non-null int64 5 tripduration 6427 non-null object 6 from_station_id 6427 non-null int64 7 from_station_name 6427 non-null object 8 to_station_id 6427 non-null int64 9 to_station_name 6427 non-null object 10 usertype 6427 non-null object 11 gender 5938 non-null object 12 birthyear 5956 non-null float6413 age 6427 non-null int32
dtypes: float64(1), int32(1), int64(5), object(7)
memory usage: 678.0+ KB#发现年龄最小是0最大是113，属于数据异常，进行数据清洗，这里保留用户年龄在18-70岁之间的群体
mobike=mobike[mobike['age']<=70]
mobike=mobike[mobike['age']>=18]
mobike.describe()user_id timeduration bikeid from_station_id to_station_id birthyear age
count 5.939000e+03 5939.000000 5939.000000 5939.000000 5939.000000 5939.000000 5939.000000
mean 2.136120e+07 10.817309 3506.519448 196.996969 201.032665 1982.591177 36.408823
std 2.182745e+05 8.477247 1916.098846 148.159023 148.888064 10.924396 10.924396
min 2.098358e+07 0.000000 2.000000 2.000000 3.000000 1949.000000 18.000000
25% 2.117430e+07 5.000000 1853.500000 77.000000 81.000000 1977.000000 28.000000
50% 2.136393e+07 8.000000 3637.000000 172.000000 174.000000 1986.000000 33.000000
75% 2.154985e+07 14.000000 5207.500000 288.000000 288.000000 1991.000000 42.000000
max 2.174223e+07 59.000000 6470.000000 662.000000 661.000000 2001.000000 70.000000#将开始时间和结束时间转变为日期时间格式，object换成datetime64
mobike['start_time']=pd.to_datetime(mobike['start_time'])
mobike['end_time']=pd.to_datetime(mobike['end_time'])
mobike.info()<class 'pandas.core.frame.DataFrame'>
Int64Index: 5939 entries, 0 to 6426
Data columns (total 14 columns):# Column Non-Null Count Dtype
--- ------ -------------- ----- 0 user_id 5939 non-null int64 1 start_time 5939 non-null datetime64[ns]2 end_time 5939 non-null datetime64[ns]3 timeduration 5939 non-null int64 4 bikeid 5939 non-null int64 5 tripduration 5939 non-null object 6 from_station_id 5939 non-null int64 7 from_station_name 5939 non-null object 8 to_station_id 5939 non-null int64 9 to_station_name 5939 non-null object 10 usertype 5939 non-null object 11 gender 5922 non-null object 12 birthyear 5939 non-null float64 13 age 5939 non-null int32
dtypes: datetime64[ns](2), float64(1), int32(1), int64(5), object(5)
memory usage: 672.8+ KBmobike.drop(mobike.select_dtypes(['datetime64']),inplace=True,axis=1)
#inplace=True直接在原始数据中保存排序之后的结果
#axis 列索引，表名不同列，纵向索引，叫columns，1轴，axis=1#把骑行事件中的逗号进行类型转换.(逗号替换掉)
mobike['tripduration'] = mobike['tripduration'].str.replace(',','').astype(int)mobike['tripduration']
0 436
1 445
2 1090
3 581
4 390...
6422 431
6423 918
6424 1165
6425 500
6426 483
Name: tripduration, Length: 5939, dtype: int32
mobike_ = mobike[['usertype','gender']]#把性别里面的缺失值用unkown代替作为填充.
mobike.gender.fillna('unkown',inplace=True)#用户渠道分类统计
mobike.groupby(["usertype"]).count() #Customer注册的，Subscriber没注册的.
pd.get_dummies(mobike_) #可以分别做哑变量转换，也可以放在一起.此时把这两列做一个哑变量矩阵#将类别性别转化成哑变量(gender)
# mobike["gender"]= pd.get_dummies(mobike.gender)
# mobike["gender"]
#将类别用户种类转化成哑变量(usertype)
# mobike["usertype"]= pd.get_dummies(mobike.usertype)
# mobike["usertype"]
mobike["gender"].value_counts() #性别个数数据统计
Male 4623
Female 1299
unkown 17
Name: gender, dtype: int64#查看当前有多少列，列名分别时什么。
mobike.columns
Index(['user_id', 'timeduration', 'bikeid', 'tripduration', 'from_station_id','from_station_name', 'to_station_id', 'to_station_name', 'usertype','gender', 'birthyear', 'age'],dtype='object')#pd.concat实现数据合并 ,传入的第一个数据是预处理的数据。
mobike_one_hot=pd.concat([mobike,pd.get_dummies(mobike_)],axis=1)
mobike_one_hot.columns
Index(['user_id', 'timeduration', 'bikeid', 'tripduration', 'from_station_id','from_station_name', 'to_station_id', 'to_station_name', 'usertype','gender', 'birthyear', 'age', 'gender_Female', 'gender_Male','gender_unkown', 'usertype_Customer', 'usertype_Subscriber'],dtype='object')
mobike_one_hot
结果展示
tripduration中的233732属于不正常的距离，需要进行数据清洗
mobike_= mobike_one_hot[['timeduration','tripduration', 'age','usertype_Customer', 'usertype_Subscriber', 'gender_Female','gender_Male', 'gender_unkown']]mobike_3=mobike_one_hot[["timeduration","tripduration","age","gender_Female","gender_Male","usertype_Customer","usertype_Subscriber"]]mobike_2= mobike_one_hot[['tripduration', 'age','usertype_Customer', 'usertype_Subscriber']]mobike.describe()user_id timeduration bikeid tripduration from_station_id to_station_id birthyear age
count 5.939000e+03 5939.000000 5939.000000 5939.000000 5939.000000 5939.000000 5939.000000 5939.000000
mean 2.136120e+07 10.817309 3506.519448 793.080485 196.996969 201.032665 1982.591177 36.408823
std 2.182745e+05 8.477247 1916.098846 3334.075550 148.159023 148.888064 10.924396 10.924396
min 2.098358e+07 0.000000 2.000000 61.000000 2.000000 3.000000 1949.000000 18.000000
25% 2.117430e+07 5.000000 1853.500000 338.500000 77.000000 81.000000 1977.000000 28.000000
50% 2.136393e+07 8.000000 3637.000000 527.000000 172.000000 174.000000 1986.000000 33.000000
75% 2.154985e+07 14.000000 5207.500000 857.000000 288.000000 288.000000 1991.000000 42.000000
max 2.174223e+07 59.000000 6470.000000 233732.000000 662.000000 661.000000 2001.000000 70.000000#由上表可知tripduration属于不正常的距离。所以把大于10000的统计之后，做删除处理。
mobike[mobike["tripduration"]>10000].count()user_id 15
timeduration 15
bikeid 15
tripduration 15
from_station_id 15
from_station_name 15
to_station_id 15
to_station_name 15
usertype 15
gender 15
birthyear 15
age 15
dtype: int64mobike=mobike[mobike['tripduration']<10000]
mobike.describe()user_id timeduration bikeid tripduration from_station_id to_station_id birthyear age
count 5.924000e+03 5924.000000 5924.000000 5924.000000 5924.000000 5924.000000 5924.000000 5924.000000
mean 2.136154e+07 10.739872 3506.339804 704.265361 197.110230 200.842674 1982.594531 36.405469
std 2.181799e+05 8.291724 1916.579026 664.976888 148.199333 148.559883 10.921172 10.921172
min 2.098358e+07 0.000000 2.000000 61.000000 2.000000 3.000000 1949.000000 18.000000
25% 2.117469e+07 5.000000 1853.000000 337.750000 77.000000 81.000000 1977.000000 28.000000
50% 2.136492e+07 8.000000 3635.500000 526.500000 173.000000 174.000000 1986.000000 33.000000
75% 2.155034e+07 14.000000 5207.250000 854.250000 288.250000 288.000000 1991.000000 42.000000
max 2.174223e+07 59.000000 6470.000000 9893.000000 662.000000 661.000000 2001.000000 70.000000
查看数据之间的相关度
#如何查看相关性 corr计算的是夹角的余弦值[-1,1].一般情况下骑得越久就越远，还是有相关性的。
mobike.corr() # 0.269336是有相关性的,0.2到0.3会有一个很明显的相关性,如果到0.3以上有明显的相关性；0.5以上有强相关性
#建模时分群tripduration和timeduration可以只用其中一个，因为用户特征比较明显；也可以两个都用。
选择七个变量mobike_3作为分群的维度，数据标准化后，进行聚类分析。
mobike_3=mobike_one_hot[["timeduration","tripduration","age","gender_Female","gender_Male","usertype_Customer","usertype_Subscriber"]]#Standardization标准化:将特征数据的分布调整成标准正太分布，也叫高斯分布，也就是使得数据的均值维0，方差为1.
#数据标准化，使用sklearn中预处理的scale
#duration=mobike_3[['tripduration','timeduration']]
#from sklearn.preprocessing import scale
#x1=pd.DataFrame(scale(duration))
#x1
mobike_3["tripduration"]=(mobike_3["tripduration"]-mobike_3["tripduration"].mean())/mobike_3["tripduration"].std()
#（收入-收入均值）/收入标准差
mobike_3["timeduration"]=(mobike_3["timeduration"]-mobike_3["timeduration"].mean())/mobike_3["timeduration"].std()#数据进行标准化后的结果，先尝试分为3类
model=KMeans(n_clusters=3,random_state=10) #n_clusters=3,分成三类，random_state自动生成随机数固定。
model.fit(mobike_3)
#提取标签，查看分类结果
mobike_3['cluster']=model.labels_ #把聚类的结果放成一列排在下面表中
mobike_3.groupby("cluster").age.count()显示结果
cluster
0 2797
1 1144
2 1998
Name: age, dtype: int64#使用groupby函数，评估各个变量维度的分群效果
mobike_3.groupby(['cluster'])['age'].describe() count mean std min 25% 50% 75% max
cluster
0 2797.0 27.701108 3.019970 18.0 26.0 28.0 30.0 32.0
1 1144.0 55.107517 5.712397 47.0 50.0 55.0 59.0 70.0
2 1998.0 37.892392 3.836275 33.0 35.0 37.0 41.0 46.0结论：从mobike_3可以看出通过年龄分成三组，18岁-32岁；33岁-46岁;47岁-70岁.结论：从mobike_3可以看出通过年龄分成三组，18岁-32岁；33岁-46岁;47岁-70岁.
模型评估与优化
from sklearn import metrics#调用sklearn的metrics库
from sklearn.preprocessing import scale
#x_cluster=model.fit_predict(mobike_3)#个体与群的距离，不能用这句来执行，因为包含了cluster.所以提取所需要的值.
x_cluster = model.fit_predict(mobike_3[['timeduration','tripduration', 'age','usertype_Customer', 'usertype_Subscriber', 'gender_Female','gender_Male']])score=metrics.silhouette_score(mobike_3[['timeduration','tripduration', 'age','usertype_Customer', 'usertype_Subscriber', 'gender_Female','gender_Male']],x_cluster)#评分越高，个体与群越近；评分越低，个体与群越远
#print(score)
#0.5399278008634186
centers=pd.DataFrame(model.cluster_centers_,columns=["timeduration","tripduration","age","gender_Female","gender_Male","usertype_Customer","usertype_Subscriber"])
centers| timeduration | tripduration | age | gender_Female | gender_Male | usertype_Customer | usertype_Subscriber | |
|---|---|---|---|---|---|---|---|
| 0 | -0.020076 | 0.006295 | 27.701108 | 0.042188 | 0.957812 | 0.258849 | 0.738649 |
| 1 | 0.048773 | 0.018423 | 55.107517 | 0.026224 | 0.973776 | 0.180070 | 0.817308 |
| 2 | 0.000178 | -0.019362 | 37.892392 | 0.030531 | 0.969469 | 0.184685 | 0.811812 |
选择四个变量mobike_2作为分群的维度，数据标准化后，进行聚类分析。
mobike_2=mobike_one_hot[["tripduration","age","usertype_Customer","usertype_Subscriber"]]
mobike_2["tripduration"]=(mobike_2["tripduration"]-mobike_2["tripduration"].mean())/mobike_2["tripduration"].std()
mobike_2["age"]=(mobike_2["age"]-mobike_2["age"].mean())/mobike_2["age"].std()#数据进行标准化后的结果，先尝试分为3类
model=KMeans(n_clusters=3,random_state=10) #n_clusters=3,分成三类，random_state自动生成随机数固定。
model.fit(mobike_2)
#提取标签，查看分类结果
mobike_2['cluster']=model.labels_ #把聚类的结果放成一列排在下面表中
mobike_2.groupby("cluster").count()显示结果tripduration age usertype_Customer usertype_Subscriber
cluster
0 4375 4375 4375 4375
1 1563 1563 1563 1563
2 1 1 1 1mobike_2["cluster"].count()
显示结果
5939模型评估与优化
x_cluster = model.fit_predict(mobike_2[["tripduration","age","usertype_Customer","usertype_Subscriber"]])
score=metrics.silhouette_score(mobike_2[["tripduration","age","usertype_Customer","usertype_Subscriber"]],x_cluster)#评分越高，个体与群越近；评分越低，个体与群越远
print(score)
0.6183295392287478centers=pd.DataFrame(model.cluster_centers_,columns=["tripduration","age","usertype_Customer","usertype_Subscriber"])
centers
| tripduration | age | usertype_Customer | usertype_Subscriber | |
|---|---|---|---|---|
| 0 | 0.005522 | 1.434627 | 0.027511 | 0.972489 |
| 1 | -0.017942 | -0.512187 | 0.037943 | 0.962057 |
| 2 | 69.866119 | -1.502035 | 0.000000 | 1.000000 |
这篇关于共享单车线性回归分析的文章就介绍到这儿，希望我们推荐的文章对编程师们有所帮助！
