用户画像的应用场景

产品层面的宏观分析维度

用户画像标签类型

聚类和分类q：

聚类的距离：

3.2 K-means算法原理

模型评估：

误差平方和

Q1：K值得确定：肘部法则

Q2: 聚类效果如何评估？？：轮廓系数

3.3 K-means算法的优缺点

3.4 使用Pyhton做K-means算法的步骤

4. 用户画像

5.使用K-means做用户画像

（1）数据解释
本节对互联网行业用户进行聚类分析，根据获取到用户的10个维度进行聚类分析，10维度涵盖了用户的观看、关注、消费行为。

KMeans(n_clusters=8, init='k-means++', n_init=10, max_iter=300, tol=0.0001,
precompute_distances='auto', verbose=0, random_state=None,
copy_x=True, n_jobs=None, algorithm='auto')

2) 聚类步骤

确定k值
建模分析
效果检测
聚类结果的解读
.

代码部分：

## 获取数据
from sklearn.metrics import r2_score
import statsmodels.api as sm
import numpy as np 
import pandas as pd 
import matplotlib.pyplot as plt
data =  pd.read_table('C:/Users/lb/Desktop/test/k-means_data.txt',sep='\t',engine="python",encoding = 'gbk')
data.columns.values
data.head()

data = data.drop(['用户id'],axis=1)print( data.shape )

缺失值

pd.isnull(data).sum()

填充数据

# 缺失值处理 -- 每列特征的均值填充缺失数据
data = data.fillna(data.mean())

#缺失值处理之后，索引恢复
for i in [data]:i.index = range(i.shape[0])
data.head()

标准化以及修正列名

from sklearn import preprocessing
scaler = preprocessing.StandardScaler().fit(data)
data_s = pd.DataFrame(scaler.fit_transform(data,y=data.columns)) 
data_s.head()

new_col = data.columns
data_s.columns = new_col
data_s.head()

标准化后的维数

data_s.shape

元数据散点图

# 查看元数据散点图
import numpy as np
import matplotlib.pyplot as plt
plt.rcParams['font.sans-serif']=['SimHei']
plt.rcParams['axes.unicode_minus'] = False
plt.scatter(data_s['浏览数量'],data_s['消费金额'], s=6, c='r', alpha=0.4, label='散点1')
plt.show()

建模：

from sklearn.cluster import KMeans
from sklearn.preprocessing import MinMaxScaler
from sklearn.metrics import silhouette_score

#  确定K值
inertia = []
for k in range(2,10):kmeans = KMeans(n_clusters = k,random_state = 1).fit(data_s)inertia.append(np.sqrt(kmeans.inertia_))
plt.plot(range(2,10),inertia,marker = 's')
plt.xlabel('$k$') #k值
plt.ylabel('$J(C_K)$')#误差平方和
#4和5

# 存放轮廓系数
Scores = []  
plt.rcParams['font.sans-serif']=['SimHei'] #显示中文标签
plt.rcParams['axes.unicode_minus']=False #用来正常显示负号
for k in range(2, 9):estimator = KMeans(n_clusters=k)  # 构造聚类器estimator.fit(data_s)Scores.append(silhouette_score(data_s, estimator.labels_, metric='euclidean'))
X = range(2, 9)
plt.xlabel('k')
plt.ylabel('轮廓系数')
plt.plot(X, Scores, 'o-')
plt.show()

## 一行表示一个行向量

calinski_harabaz_score：卡林斯基-哈拉巴斯指数

# calinski_harabaz指数
from sklearn.metrics import calinski_harabaz_score
import matplotlib.pyplot as plt
haraba = []  
plt.rcParams['font.sans-serif']=['SimHei'] #显示中文标签
plt.rcParams['axes.unicode_minus']=False #用来正常显示负号
for k in range(2, 9):estimator = KMeans(n_clusters=k)  # 构造聚类器estimator.fit(data_s)haraba.append(calinski_harabaz_score(data_s,estimator.labels_))
X = range(2, 9)
plt.xlabel('k')
plt.ylabel('calinski_harabaz指数')
plt.plot(X, haraba, 'o-')
plt.show()

建立模型

# 建立模型
cluster = KMeans(n_clusters = 3,random_state = 5).fit(data_s)

聚类中心

# 聚类中心
centers = cluster.cluster_centers_ # 聚类中心
centers

因为是3个簇 所有是3个

样本类别标签

# 查看总距离平方和 
inertia = cluster.inertia_
inertia 

# 轮廓系数均值
silhouette_score(data_s,c_preds1) 

# 轮廓系数
from sklearn.metrics import silhouette_score
from sklearn.metrics import silhouette_samples
#每个样本的轮廓系数
silhouette_samples(data_s,c_preds1)

array([ 0.92732185, 0.93211408, 0.91754354, 0.88674901, 0.93250814,
0.91832453, 0.93225057, 0.91800177, 0.91801442, 0.91371574,
0.93039028, 0.92780645, 0.9175957 , 0.91805761, 0.91801695,
0.90394838, 0.82497174, 0.91733802, 0.86933501, 0.92838568,
0.90395603, 0.90396898, 0.71949046, 0.60180808, 0.91797249,
0.90003841, 0.8249717 , 0.91807368, 0.93202087, 0.9181117 ,
0.93245682, 0.92778811, 0.92656082, 0.72477047, 0.93019856,
0.92828162, 0.91816004, 0.88759818, 0.93127829, 0.92603308,
0.90071358, -0.17466512, 0.93172762, 0.91827221, 0.8707087 ,
0.93098797, 0.93098797, 0.93137879, 0.93064253, 0.90396551,
0.91799883, 0.93294867, 0.93255235, 0.92731491, 0.91761448,
0.91798014, 0.93159375, 0.91974077, 0.91644496, 0.85981065,
0.90395365, 0.91817525, 0.9318612 , 0.90394838, 0.91843222,
0.61614305, 0.93230046, 0.93227226, 0.93062299, 0.91772833,
0.92723017, 0.91801529, 0.91801117, 0.93302561, 0.89769678,
0.91819631, 0.90394838, 0.92225906, 0.42998441, 0.85359922,
0.49967657, 0.35952168, 0.91810681, 0.92582604, 0.9182524 ,
0.66910184, 0.91805175, 0.91805175, 0.93255546, 0.91548173,
0.91548173, 0.91815125, 0.37411391, 0.60619449, 0.91929334,
0.93075391, 0.40382763, 0.9278305 , 0.93120899, 0.93308297,
0.93150594, 0.93177713, 0.90394982, 0.90394982, 0.91842709,
0.92716037, 0.92970515, 0.93027568, 0.93142373, 0.93067922,
0.93228489, 0.91605913, 0.8630899 , 0.92845676, 0.91807434,
0.56483434, 0.59047301, 0.93211681, 0.93195585, 0.91769925,
0.93239282, 0.91810854, 0.93265678, 0.52170789, 0.91768007,
0.79704101, 0.9174981 , 0.93153993, 0.9273972 , 0.91796437,
0.93145531, 0.93198917, 0.90394888, 0.18277362, 0.91824065,
0.91754479, 0.91798471, 0.9317068 , 0.9301013 , 0.91741988,
0.93116962, 0.03740003, 0.91805067, 0.93212341, 0.30722591,
0.47547937, 0.92527144, 0.6330433 , 0.6330433 , 0.9310836 ,
0.87741024, 0.90396363, 0.91814416, 0.91631695, 0.34389833,
0.93233374, 0.91748558, 0.93103558, 0.50049229, 0.91844024,
0.8911124 , 0.8911124 , 0.91857703, 0.91148529, 0.93050201,
0.90395961, 0.91174891, 0.90395527, 0.93252806, 0.91797483,
0.93297992, 0.92870732, 0.93023766, 0.83658348, 0.92388233,
0.93170759, 0.93160029, 0.93270814, 0.4490633 , 0.4490633 ,
0.78738663, 0.92915066, 0.9268259 , 0.93305011, 0.9280447 ,
0.91796439, 0.91738324, 0.75867505, 0.92721804, 0.91841875,
0.91957815, 0.91844261, 0.93167406, 0.91834179, 0.9180343 ,
0.92740537, 0.9174482 , 0.92530647, 0.69268396, 0.91137483,
0.91808896, 0.9279222 , 0.93057453, 0.8968111 , 0.91815978,
0.82378692, 0.83671792, 0.91813035, 0.91796654, 0.91805443,
0.92863594, 0.93098307, 0.88998633, 0.92986096, 0.9236901 ,
0.91805397])

一共是216个样本 ，所以是216个轮廓系数

len( silhouette_samples(data_s,c_preds1) )

取其中两列 画聚类中心图 因为有三个簇是3个

import matplotlib.pyplot as plt 
plt.figure()
plt.scatter(data_s.values[:, 0], data_s.values[:, 1], c=c_preds1)#原始数据散点图，按照分类查看
plt.scatter(centers[:, 0], centers[:, 1],marker='x', s=169, linewidths=3,color='r', zorder=10)#重心红色X进行突出

聚类和标签组合起来

clust_prod = data.copy()
clust_prod['标签'] = c_preds1
clust_prod.head(10)

clust_prod[clust_prod.标签 == 0]

直方图观察各类中某个变量的区别

# 绘制子图
import matplotlib.pyplot as plt
#fig = plt.figure()
for i in range(3):plt.subplot(1,3,i+1)plt.hist(x = clust_prod[clust_prod.标签==i].浏览数量, # 指定绘图数据bins = 5, # 指定直方图中条块的个数color = 'steelblue', # 指定直方图的填充色edgecolor = 'black' # 指定直方图的边框色)plt.title('第'+str(i)+'类')
plt.show()

层次聚类：

AGNES，是一种采用自底向上聚合策略的层次聚类算法，是先将数据集中的每个样本看作一个初始聚类簇，然后在算法运行的每一步中找出距离最近的两个聚类簇进行合并，该过程不断重复，直至达到预设的聚类簇个数，这里的关键是如何计算聚类簇之间的距离。其实每个聚类簇就是一个样本集合，聚类簇之间的聚类就是集合之间的聚类。

from sklearn.cluster import AgglomerativeClustering
from sklearn.metrics import confusion_matrix
clustering = AgglomerativeClustering(linkage='ward', n_clusters=3)
res = clustering.fit(data_s)

es.labels_r

array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 1, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0,
2, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], dtype=int64)

# 存放轮廓系数
Scores = []  
plt.rcParams['font.sans-serif']=['SimHei'] #显示中文标签
plt.rcParams['axes.unicode_minus']=False #用来正常显示负号
for k in range(2, 9):estimator = AgglomerativeClustering(n_clusters=k)  # 构造聚类器estimator.fit(data_s)Scores.append(silhouette_score(data_s, estimator.labels_, metric='euclidean'))
X = range(2, 9)
plt.xlabel('k')
plt.ylabel('轮廓系数')
plt.plot(X, Scores, 'o-')
plt.show()