Python用K-means聚类算法进行客户分群的实现

import numpy as np
import pandas as pd

from pandas import plotting
import matplotlib.pyplot as plt
import seaborn as sns
import plotly.graph_objs as go
import plotly.offline as py

from sklearn.cluster import KMeans

import warnings
warnings.filterwarnings('ignore')

io = '.../Mall_Customers.csv'
df = pd.DataFrame(pd.read_csv(io))
# 修改列名
df.rename(columns={'Annual Income (k$)': 'Annual Income', 'Spending Score (1-100)': 'Spending Score'}, inplace=True)
print(df.head())
print(df.describe())
print(df.shape)
print(df.count())
print(df.dtypes)

plotting.parallel_coordinates(df.drop('CustomerID', axis=1), 'Gender')
plt.title('平行坐标图', fontsize=12)
plt.grid(linestyle='-.')
plt.show()

sns.set(palette="muted", color_codes=True)  # seaborn样式
# 配置
plt.rcParams['axes.unicode_minus'] = False  # 解决无法显示符号的问题
sns.set(font='SimHei', font_scale=0.8)    # 解决Seaborn中文显示问题
# 绘图
plt.figure(1, figsize=(13, 6))
n = 0
for x in ['Age', 'Annual Income', 'Spending Score']:
  n += 1
  plt.subplot(1, 3, n)
  plt.subplots_adjust(hspace=0.5, wspace=0.5)
  sns.distplot(df[x], bins=16, kde=True)  # kde 密度曲线
  plt.title('{}分布情况'.format(x))
  plt.tight_layout()
plt.show()

plt.figure(1, figsize=(13, 6))
k = 0
for x in ['Age', 'Annual Income', 'Spending Score']:
  k += 1
  plt.subplot(3, 1, k)
  plt.subplots_adjust(hspace=0.5, wspace=0.5)
  sns.countplot(df[x], palette='rainbow', alpha=0.8)
  plt.title('{}分布情况'.format(x))
  plt.tight_layout()
plt.show()

df_gender_c = df['Gender'].value_counts()
p_lables = ['Female', 'Male']
p_color = ['lightcoral', 'lightskyblue']
p_explode = [0, 0.05]
# 绘图
plt.pie(df_gender_c, labels=p_lables, colors=p_color, explode=p_explode, shadow=True, autopct='%.2f%%')
plt.axis('off')
plt.legend()
plt.show()

# df_a_a_s = df.drop(['CustomerID'], axis=1)
sns.pairplot(df, vars=['Age', 'Annual Income', 'Spending Score'], hue='Gender', aspect=1.5, kind='reg')
plt.show()

# 根据分类变量分组绘制一个纵向的增强箱型图
plt.rcParams['axes.unicode_minus'] = False  # 解决无法显示符号的问题
sns.set(font='SimHei', font_scale=0.8)    # 解决Seaborn中文显示问题
sns.boxenplot(df['Gender'], df['Spending Score'], palette='Blues')
# x:设置分组统计字段，y:数据分布统计字段
sns.swarmplot(x=df['Gender'], y=df['Spending Score'], data=df, palette='dark', alpha=0.5, size=6)
plt.title('男女性的消费能力比较', fontsize=12)
plt.show()

# 根据分类变量分组绘制一个纵向的增强箱型图
plt.rcParams['axes.unicode_minus'] = False  # 解决无法显示符号的问题
sns.set(font='SimHei', font_scale=0.8)    # 解决Seaborn中文显示问题
sns.boxenplot(df['Gender'], df['Spending Score'], palette='Blues')
# x:设置分组统计字段，y:数据分布统计字段
sns.swarmplot(x=df['Gender'], y=df['Spending Score'], data=df, palette='dark', alpha=0.5, size=6)
plt.title('男女性的消费能力比较', fontsize=12)
plt.show()

df_a_sc = df[['Age', 'Spending Score']].values
# 存放每次聚类结果的误差平方和
inertia1 = []

for n in range(1, 11):
  # 构造聚类器
  km1 = (KMeans(n_clusters=n,    # 要分成的簇数，int类型，默认值为8
         init='k-means++',  # 初始化质心，k-means++是一种生成初始质心的算法
         n_init=10,      # 设置选择质心种子次数，默认为10次。返回质心最好的一次结果（好是指计算时长短）
         max_iter=300,    # 每次迭代的最大次数
         tol=0.0001,     # 容忍的最小误差，当误差小于tol就会退出迭代
         random_state=111,  # 随机生成器的种子 ，和初始化中心有关
         algorithm='elkan')) # 'full'是传统的K-Means算法，'elkan'是采用elkan K-Means算法
  # 用训练数据拟合聚类器模型
  km1.fit(df_a_sc)
  # 获取聚类标签
  inertia1.append(km1.inertia_)

plt.figure(1, figsize=(15, 6))
plt.plot(np.arange(1, 11), inertia1, 'o')
plt.plot(np.arange(1, 11), inertia1, '-', alpha=0.7)
plt.title('手肘法图', fontsize=12)
plt.xlabel('聚类数'), plt.ylabel('SSE')
plt.grid(linestyle='-.')
plt.show()

km1_result = (KMeans(n_clusters=4, init='k-means++', n_init=10, max_iter=300,
           tol=0.0001, random_state=111, algorithm='elkan'))
# 先fit()再predict()，一次性得到聚类预测之后的标签
y1_means = km1_result.fit_predict(df_a_sc)
# 绘制结果图
plt.scatter(df_a_sc[y1_means == 0][:, 0], df_a_sc[y1_means == 0][:, 1], s=70, c='blue', label='1', alpha=0.6)
plt.scatter(df_a_sc[y1_means == 1][:, 0], df_a_sc[y1_means == 1][:, 1], s=70, c='orange', label='2', alpha=0.6)
plt.scatter(df_a_sc[y1_means == 2][:, 0], df_a_sc[y1_means == 2][:, 1], s=70, c='pink', label='3', alpha=0.6)
plt.scatter(df_a_sc[y1_means == 3][:, 0], df_a_sc[y1_means == 3][:, 1], s=70, c='purple', label='4', alpha=0.6)
plt.scatter(km1_result.cluster_centers_[:, 0], km1_result.cluster_centers_[:, 1], s=260, c='gold', label='质心')
plt.title('聚类图(K=4)', fontsize=12)
plt.xlabel('年收入(k$)')
plt.ylabel('消费分数(1-100)')
plt.legend()
plt.grid(linestyle='-.')
plt.show()

df_ai_sc = df[['Annual Income', 'Spending Score']].values
# 存放每次聚类结果的误差平方和
inertia2 = []

for n in range(1, 11):
  # 构造聚类器
  km2 = (KMeans(n_clusters=n, init='k-means++', n_init=10, max_iter=300, tol=0.0001, random_state=111, algorithm='elkan'))
  # 用训练数据拟合聚类器模型
  km2.fit(df_ai_sc)
  # 获取聚类标签
  inertia2.append(km2.inertia_)
# 绘制手肘图确定K值
plt.figure(1, figsize=(15, 6))
plt.plot(np.arange(1, 11), inertia1, 'o')
plt.plot(np.arange(1, 11), inertia1, '-', alpha=0.7)
plt.title('手肘法图', fontsize=12)
plt.xlabel('聚类数'), plt.ylabel('SSE')
plt.grid(linestyle='-.')
plt.show()

km2_result = (KMeans(n_clusters=5, init='k-means++', n_init=10, max_iter=300,
           tol=0.0001, random_state=111, algorithm='elkan'))
# 先fit()再predict()，一次性得到聚类预测之后的标签
y2_means = km2_result.fit_predict(df_ai_sc)
# 绘制结果图
plt.scatter(df_ai_sc[y2_means == 0][:, 0], df_ai_sc[y2_means == 0][:, 1], s=70, c='blue', label='1', alpha=0.6)
plt.scatter(df_ai_sc[y2_means == 1][:, 0], df_ai_sc[y2_means == 1][:, 1], s=70, c='orange', label='2', alpha=0.6)
plt.scatter(df_ai_sc[y2_means == 2][:, 0], df_ai_sc[y2_means == 2][:, 1], s=70, c='pink', label='3', alpha=0.6)
plt.scatter(df_ai_sc[y2_means == 3][:, 0], df_ai_sc[y2_means == 3][:, 1], s=70, c='purple', label='4', alpha=0.6)
plt.scatter(df_ai_sc[y2_means == 4][:, 0], df_ai_sc[y2_means == 4][:, 1], s=70, c='green', label='5', alpha=0.6)
plt.scatter(km2_result.cluster_centers_[:, 0], km2_result.cluster_centers_[:, 1], s=260, c='gold', label='质心')
plt.title('聚类图(K=5)', fontsize=12)
plt.xlabel('年收入(k$)')
plt.ylabel('消费分数(1-100)')
plt.legend()
plt.grid(linestyle='-.')
plt.show()

df_a_ai_sc = df[['Age', 'Annual Income', 'Spending Score']].values

km3 = KMeans(n_clusters=5, init='k-means++', max_iter=300, n_init=10, random_state=0)
km3.fit(df_a_ai_sc)

df['labels'] = km3.labels_
# 绘制3D图
trace1 = go.Scatter3d(
  x=df['Age'],
  y=df['Spending Score'],
  z=df['Annual Income'],
  mode='markers',
   marker=dict(
    color=df['labels'],
    size=10,
    line=dict(
      color=df['labels'],
      width=12
    ),
    opacity=0.8
   )
)
df_3dfid = [trace1]

layout = go.Layout(
  margin=dict(
    l=0,
    r=0,
    b=0,
    t=0
  ),
  scene=dict(
      xaxis=dict(title='年龄'),
      yaxis=dict(title='消费分数(1-100)'),
      zaxis=dict(title='年收入(k$)')
    )
)

fig = go.Figure(data=df_3dfid, layout=layout)
py.offline.plot(fig)