(数据挖掘 —— 无监督学习（聚类）

1. K-means

K-Means为基于切割的聚类算法

1.1 生成指定形状的随机数据

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from sklearn.cluster import KMeans
# *************** 生成指定形状的随机数据 *****************
from sklearn.datasets import make_circles,make_moons,make_blobs
n_samples = 1000

# 生成环装数据
circles = make_circles(n_samples = n_samples,factor = 0.5,noise = 0.05)
"""
n_samples: 为样本点个数
factor:为大圆与小圆的间距
"""
# 生成月牙形数据
moons = make_moons(n_samples = n_samples,noise = 0.05)

# 生成簇状数据
blobs = make_blobs(n_samples = n_samples,random_state = 100,center_box = (-10,10),cluster_std = 1,centers = 3)
"""
random_state: 随机数种子，多少代保持随机数不变
center_box: 中心确定后的数据边界 默认(-10,10)
cluster_std：数据分布的标准差，决定各类数据的紧凑程度,默认为1.0
centers：产生数据点中心的个数 默认为3
"""
# 产生随机数
random_data = np.random.rand(n_samples,2),np.array([0 for i in range(n_samples)])
datasets = [circles,moons,blobs,random_data]
fig = plt.figure(figsize=(20,8))

1.2 进行聚类

colors = "rgbykcm"
for index,data in enumerate(datasets):
    X = data[0]
    Y_old = data[1]
    km_cluster = KMeans(n_clusters = 2)
    km_cluster.fit(X)
    Y_new = km_cluster.labels_
    fig.add_subplot(2,len(datasets),index+1)
    [plt.scatter(X[i,0],X[i,1],color = colors[Y_old[i]]) for i in range(len(X[:,0]))] 
    fig.add_subplot(2,len(datasets),index+5)
    [plt.scatter(X[i,0],X[i,1],color = colors[Y_new[i]]) for i in range(len(X[:,0]))]

1.3 结果

2. 系统聚类

2.1 代码

AgglomerativeClustering(n_clusters,affinity,linkage)

• affinity:

1. “euclidean”,欧几里得距离
2. “l1”, “l2”,
3. “manhattan”, 曼哈顿距离
4. “cosine”, 余弦距离
5. “precomputed”预输入 需要输出距离矩阵

• linkage：{“ward”, “complete”, “average”, “single”}, default=”ward”

from sklearn.datasets import make_circles,make_blobs,make_moons
from sklearn.cluster import AgglomerativeClustering
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd

# 准备数据
n_samples = int(1e3)
circles = make_circles(n_samples = n_samples,noise = 0.05,factor = 0.5,random_state = 10)
moons = make_moons(n_samples = n_samples,noise = 0.05,random_state = 10)
blobs = make_blobs(n_samples=n_samples,centers = 4,cluster_std = 0.1,center_box = (-1,1),random_state = 10)
np.random.seed(10)
random_data = (np.random.rand(n_samples,2),np.zeros((n_samples)).astype(np.int))

datasets = [circles,moons,blobs,random_data]
fig = plt.figure(figsize = (20,8),dpi = 72)
colors = "rgbk"
for index,data in enumerate(datasets):
    X = data[0]
    Y = data[1]
    agg_cluster = AgglomerativeClustering(n_clusters = 2,affinity = "euclidean",linkage = "average")
    Y_predict = agg_cluster.fit(X).labels_
    fig.add_subplot(2,len(datasets),index + 1)
    [plt.scatter(X[i,0],X[i,1],color = colors[Y[i]]) for i in range(len(X[:,0]))]
    fig.add_subplot(2,len(datasets),index + 5)
    [plt.scatter(X[i,0],X[i,1],color = colors[Y_predict[i]]) for i in range(len(X[:,0]))]
    

2.2 结果

3 DBSCAN

3.1 参数选择

1. 半径：k距离帮助设置半径，也就是要找到突变点，
   即选中一个点，计算它和所有其他点的距离，
   从小到大排序，发现距离突变点。
   需要做大量实验观察。
2. MinPts：先设置偏小一些，然后进行多次尝试

3.2 代码

# 导入聚类数据
n_samples = 1000
from sklearn.datasets import make_circles,make_moons,make_blobs
from sklearn.cluster import DBSCAN
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
circles = make_circles(n_samples = n_samples,noise = 0.05,factor = 0.5,random_state = 10)
moons = make_moons(n_samples = n_samples,noise = 0.05,random_state = 10)
blobs = make_blobs(n_samples = n_samples,centers = 3,cluster_std = 0.1,center_box = (-1,1),random_state = 10)
np.random.seed(10)
random_data = (np.random.rand(n_samples,2),np.zeros((n_samples)).astype(np.int))
datasets = [circles,moons,blobs,random_data]
fig = plt.figure(figsize = (20,8),dpi = 72)
colors = "rgbky"
for index,data in enumerate(datasets):
    X = data[0]
    Y_old = data[1]
    dbscan_model = DBSCAN(eps = 0.1,min_samples = 20)
    dbscan_model.fit(X)
    Y_new = dbscan_model.labels_
    fig.add_subplot(2,len(datasets),index+1)
    [plt.scatter(X[i,0],X[i,1],color = colors[Y_old[i]]) for i in range(len(X[:,0]))]
    plt.title("original algorithm")
    fig.add_subplot(2,len(datasets),index + 5)
    [plt.scatter(X[i,0],X[i,1],color = colors[Y_new[i]]) for i in range(len(X[:,0]))]
    plt.title("DBSCA algorithm")

3.3 结果

by CyrusMay 2022 04 05