《Python数据分析与挖掘实战》第7章

实战》第7章"/>

《Python数据分析与挖掘实战》第7章

本文是基于《Python数据分析与挖掘实战》的第七章的数据——《航空公司客户价值分析》做的分析对部分代码，做出补充，对原文中的雷达图进行了实现。

1. 背景与目标分析

此项目旨在根据航空公司提供的数据，对其客户进行分类，并且比较不同类别客户的价值，为能够更好的为客户提供个性化服务做参考。

2. 整体流程如下

import pandas as pd
import numpy as np

data = pd.read_csv('air_data.csv', encoding='utf8')

data.head()

5 rows × 44 columns

import chardetf = open('air_data.csv', 'rb')
data_1 = f.read()
print(chardet.detect(data_1))

{'encoding': 'utf-8', 'confidence': 0.99, 'language': ''}

print(type(data))
# print(data.head())
data.isnull().sum()

<class 'pandas.core.frame.DataFrame'>MEMBER_NO                     0
FFP_DATE                      0
FIRST_FLIGHT_DATE             0
GENDER                        3
FFP_TIER                      0
WORK_CITY                  2269
WORK_PROVINCE              3248
WORK_COUNTRY                 26
AGE                         420
LOAD_TIME                     0
FLIGHT_COUNT                  0
BP_SUM                        0
EP_SUM_YR_1                   0
EP_SUM_YR_2                   0
SUM_YR_1                    551
SUM_YR_2                    138
SEG_KM_SUM                    0
WEIGHTED_SEG_KM               0
LAST_FLIGHT_DATE              0
AVG_FLIGHT_COUNT              0
AVG_BP_SUM                    0
BEGIN_TO_FIRST                0
LAST_TO_END                   0
AVG_INTERVAL                  0
MAX_INTERVAL                  0
ADD_POINTS_SUM_YR_1           0
ADD_POINTS_SUM_YR_2           0
EXCHANGE_COUNT                0
avg_discount                  0
P1Y_Flight_Count              0
L1Y_Flight_Count              0
P1Y_BP_SUM                    0
L1Y_BP_SUM                    0
EP_SUM                        0
ADD_Point_SUM                 0
Eli_Add_Point_Sum             0
L1Y_ELi_Add_Points            0
Points_Sum                    0
L1Y_Points_Sum                0
Ration_L1Y_Flight_Count       0
Ration_P1Y_Flight_Count       0
Ration_P1Y_BPS                0
Ration_L1Y_BPS                0
Point_NotFlight               0
dtype: int64

data.describe()

8 rows × 36 columns

data = data[data['SUM_YR_1'].notnull() & data['SUM_YR_2'].notnull()]
data_1 = data['SUM_YR_1'] != 0
data_2 = data['SUM_YR_2'] != 0
data_3 = (data['SEG_KM_SUM'] == 0) & (data['avg_discount'] == 0)
data = data[data_1 | data_2 | data_3] #该规则是“或”

data.head()
data.describe()

8 rows × 36 columns

data.to_csv('air_data_cleaned.csv') # 导出结果

data1 = pd.DataFrame(columns=['R', 'L', 'F', 'M', 'C'])
# b['time_interval']=pd.to_datetime(b['xxx'])-pd.to_datetime(b['xxx'])
data1['L'] = pd.to_datetime(data.LOAD_TIME)-pd.to_datetime(data.FFP_DATE)
data1['R'] = data.LAST_TO_END
data1['M'] = data.SEG_KM_SUM
data1['C'] = data.avg_discount
data1['F'] = data.FLIGHT_COUNT

data1.info()
data1['L'] = data1['L']/np.timedelta64(1, 'D') # 将日期转化为数值，单位为天

<class 'pandas.core.frame.DataFrame'>
Int64Index: 62044 entries, 0 to 62978
Data columns (total 5 columns):
R    62044 non-null int64
L    62044 non-null timedelta64[ns]
F    62044 non-null int64
M    62044 non-null int64
C    62044 non-null float64
dtypes: float64(1), int64(3), timedelta64[ns](1)
memory usage: 2.8 MB

data2 = (data1-data1.mean(axis=0))/(data1.std(axis=0)) #标准差标准化 

from sklearn.cluster import KMeans

k=5
kmodel = KMeans(n_clusters = k, n_jobs=4)
kmodel.fit(data2)print(kmodel.cluster_centers_)
print(kmodel.labels_)

[[-0.37722119  1.16066672 -0.08691852 -0.09484404 -0.1559046 ][ 1.68625847 -0.31367829 -0.57401599 -0.53682019 -0.1733261 ][-0.41488827 -0.70020646 -0.16114258 -0.16095751 -0.25513154][-0.00266813  0.05184279 -0.22680311 -0.23125407  2.19134701][-0.79938326  0.48332845  2.4832016   2.42472391  0.30863003]]
[4 4 4 ... 2 1 1]

labels = kmodel.labels_
df1 = pd.DataFrame(labels, columns = ['numbers'])
df2 = pd.DataFrame(kmodel.cluster_centers_, columns=data1.columns)
df3 = df1['numbers'].value_counts()
df4 = pd.concat([df3, df2], axis=1)
# df4

df4

import numpy as np
import matplotlib.pyplot as plt
from matplotlib import font_manager
my_font = font_manager.FontProperties(fname="C:\Windows\Fonts\simhei.ttf")def plot_radar(data):'''the first column is the number of each cluster;the last are those to describe the center of each cluster.'''kinds = list(df4.index)kinds1=['客户群1', '客户群2', '客户群3', '客户群4', '客户群5']labels = data.iloc[:, 1:].columnssam = ['r-.', 'o-.', 'g--', 'b-', 'p:'] # 样式centers = pd.concat([data.iloc[:, 1:], data.iloc[:,1]], axis=1)  # 形成闭环centers = np.array(centers)n = len(labels)angles = np.linspace(0, 2*np.pi, n, endpoint=False)angles = np.concatenate((angles, [angles[0]])) fig = plt.figure(figsize=(7,7),dpi=130)ax = fig.add_subplot(111, polar=True) # 设置坐标为极坐标# 画若干个五边形floor = np.floor(centers.min())     # 大于最小值的最大整数ceil = np.ceil(centers.max())       # 小于最大值的最小整数for i in np.arange(floor, ceil + 0.5, 0.5):ax.plot(angles, [i] * (n + 1), '--', lw=0.5 , color='black')# 画不同客户群的分割线for i in range(n):ax.plot([angles[i], angles[i]], [floor, ceil], '--', lw=0.5, color='black')# 画不同的客户群所占的大小for i in range(len(kinds)):ax.plot(angles, centers[i], sam[i], lw=2, label=kinds1[i])ax.fill(angles, centers[i], alpha=0.25)  # ax.set_thetagrids(angles * 180 / np.pi, labels) # 设置显示的角度，将弧度转换为角度plt.title('客户群特征分布图', font_properties=my_font) # 添加标题plt.legend(prop=my_font, loc='lower right', bbox_to_anchor=(1.5, 0.0)) # 设置图例的位置，在画布外ax.set_theta_zero_location('N')        # 设置极坐标的起点（即0°）在正北方向，即相当于坐标轴逆时针旋转90°ax.spines['polar'].set_visible(False)  # 不显示极坐标最外圈的圆ax.set_ylim(floor,ceil) # 设置雷达图的范围 不设置中心就变成了五星为中心ax.grid(False) # 不显示默认的分割线ax.set_yticks( [i for i in np.arange(floor, ceil, 0.5)])                      # 显示坐标间隔plt.show()

plot_radar(df4)