机器学习4 kmeans聚类成3个簇

基操（kmeans建模，拟合，质心，label，inertia）

数据格式（左下）和最终结果（右上）

模型准备和拟合

clf = KMeans(n_clusters=3)
clf.fit(data)

查看质心cluster_centers_

clf.cluster_centers_

查看每个点所属于的簇的标签labels_

clf.labels_

查看误差平方和 inertia_

clf.inertia_

绘制点图

点图plt.scatter(横坐标对应数据列，纵坐标对应数据列，颜色c=clf.labels_,标记marker=‘x’)

plt.scatter(data['每分钟助攻数'],data['每分钟得分数'],c=clf.labels_,marker='x')

标记质心 cluster_centers_ 结合刚才已经求得的质心坐标，第1列（索引0）为横坐标，第2列（索引1）为纵坐标，
eg:只要索引为0的第一列，表示为 [:，0]，：表示行不要

plt.scatter(clf.cluster_centers_[:,0],clf.cluster_centers_[:,1],color='red')

最后加上横纵坐标，标题 xlabel,ylabel,title

横纵坐标为中文jupyter识别不了，改为英文解决

plt.scatter(data['每分钟助攻数'],data['每分钟得分数'],c=clf.labels_,marker='x')
plt.scatter(clf.cluster_centers_[:,0],clf.cluster_centers_[:,1],c='red')

plt.xlabel('assists per min')
plt.xlabel('points per min')
plt.title('KMeans Basketball data')

！！！注意，color用c肯定不会错，用color有时会报错

全文代码

import pandas as pd
from sklearn.cluster import KMeans
from matplotlib import pyplot as plt

1、准备数据集

data = pd.read_csv(r'basketball.csv')
data.head()

data.shape

2、Kmeans聚类

# 1.准备模型
km = KMeans(n_clusters=3)

# 2.训练模型
km.fit(data)

3、查看属性

# 1.质心
km.cluster_centers_

# 2.属于哪个簇
km.labels_

# 3.误差平方和
km.inertia_

4、最优模型

result = []
for n_clusters in range(2,10):
    for max_iter in  range(300,601,50):
        for tol in range(2,10):
            tol=tol*1e-5
            km = KMeans(n_clusters=n_clusters, max_iter=max_iter, tol=tol)
            km.fit(data)
            d = {'n_clusters':n_clusters, 'max_iter':max_iter, 'tol':tol, 'inertia': km.inertia_}
            result.append(d)

result

result_df = pd.DataFrame(result)

result_df

找到误差平方和最小，对应的模型参数

r = result_df.iloc[result_df['inertia'].argmin(), :]
type(r)

r

n_clusters = r['n_clusters']
n_clusters

max_iter = int(r['max_iter'])
max_iter

tol = r['tol']
tol

选择合适的K值

data.head()

data.shape

for n_clusters in range(2,20):
    print(n_clusters)

x = []
y = []
for n_clusters in range(2,20):
    print(n_clusters)
    km = KMeans(n_clusters=n_clusters, max_iter=max_iter, tol=tol)
    km.fit(data)
    # 保存n_clusters、误差平方和
    x.append(n_clusters)
    y.append(km.inertia_)

x

y

用n_clusters（k）作为横坐标，误差平方和作为中坐标---折线图（最优k）

plt.figure(figsize=(20, 8), dpi=100)
plt.plot(x, y)
plt.xlabel('K')
plt.ylabel('inertia')
plt.title('k-inertia relation')
# 将数字标注到折线图中
for a, b in zip(x, y):
    plt.text(a, b, round(b, 2), ha='center', va='bottom')

ppt-K-Means代码示例

法1

min_inertia = 10000000
k = 0

for i in range(5,30):
    km = KMeans(n_clusters=i)
    km.fit(data)
    inertia = km.inertia_
    if min_inertia > inertia:
        min_inertia = inertia
        k = i

min_inertia

k

法2

d = {}

for i in range(5,30):
    km = KMeans(n_clusters=i).fit(data)
    d[i] = km.inertia_

d

d_new = pd.Series(d)
d_new

d_new = pd.Series(d)
d_new.idxmin()

min(zip(d.values(), d.keys()))[1]

min(d.items(), key=lambda x: x[1])[0]

d_new

d_new.argmin()

d_new.argmin()

法3

e_min = 1000000
k = 0

for i in d:
#     print(i)
    if d[i] < e_min:
        e_min = d[i]
        k = i

e_min

k


### 聚类成3个簇

clf = KMeans(n_clusters=3)
clf.fit(data)

data.head()

clf.labels_

#y_pred=clf.predict(data)
#y_pred

plt.scatter(data['每分钟助攻数'],data['每分钟得分数'],c=clf.labels_,marker='x')

clf.cluster_centers_

plt.scatter(data['每分钟助攻数'],data['每分钟得分数'],c=clf.labels_,marker='x')
plt.scatter(clf.cluster_centers_[:,0],clf.cluster_centers_[:,1],c='red')

plt.xlabel('assists per min')
plt.xlabel('points per min')
plt.title('KMeans Basketball data')