七天学完十大机器学习经典算法-08.K均值聚类：无监督学习的万能分箱术

# K均值伪代码
def k_means(data, K, max_iters=100):
    # 1. 随机初始化质心
    centroids = random_select(data, K)
    
    for _ in range(max_iters):
        clusters = [[] for _ in range(K)]
        
        # 2. 分配点到最近质心
        for point in data:
            distances = [distance(point, centroid) for centroid in centroids]
            cluster_idx = argmin(distances)
            clusters[cluster_idx].append(point)
        
        old_centroids = centroids.copy()
        
        # 3. 重新计算质心
        for i in range(K):
            if clusters[i]:  # 避免空簇
                centroids[i] = mean(clusters[i])
        
        # 4. 检查收敛
        if distance(old_centroids, centroids) < threshold:
            break
    
    return clusters, centroids

# 肘部法则Python实现
from sklearn.cluster import KMeans
import matplotlib.pyplot as plt

inertias = []
k_values = range(1, 10)

for k in k_values:
    kmeans = KMeans(n_clusters=k, random_state=42)
    kmeans.fit(data)
    inertias.append(kmeans.inertia_)

plt.plot(k_values, inertias, 'bo-')
plt.xlabel('K值')
plt.ylabel('惯性(Inertia)')
plt.title('肘部法则选择最佳K值')
plt.axvline(x=3, color='r', linestyle='--')  # 假设肘点在K=3
plt.show()

# 轮廓系数评估
from sklearn.metrics import silhouette_score

best_k = 0
best_score = -1

for k in range(2, 10):
    kmeans = KMeans(n_clusters=k, random_state=42)
    labels = kmeans.fit_predict(data)
    score = silhouette_score(data, labels)
    
    if score > best_score:
        best_score = score
        best_k = k

print(f"最佳K值: {best_k}, 轮廓系数: {best_score:.4f}")

# 使用K均值++初始化
kmeans = KMeans(n_clusters=3, init='k-means++', random_state=42)

from sklearn.preprocessing import StandardScaler

scaler = StandardScaler()
data_scaled = scaler.fit_transform(data)
kmeans.fit(data_scaled)

import pandas as pd
import seaborn as sns
from sklearn.cluster import KMeans

# 加载数据
customers = pd.read_csv('customer_data.csv')

# 特征选择
X = customers[['frequency', 'avg_spent', 'recency']]

# 特征标准化
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)

# K均值聚类
kmeans = KMeans(n_clusters=3, init='k-means++', n_init=10, random_state=42)
customers['segment'] = kmeans.fit_predict(X_scaled)

# 分析聚类结果
segment_profile = customers.groupby('segment').agg({
    'frequency': 'mean',
    'avg_spent': 'mean',
    'recency': 'mean',
    'customer_id': 'count'
}).rename(columns={'customer_id': 'count'})

print(segment_profile)

# 可视化
sns.pairplot(customers, hue='segment', palette='viridis')
plt.suptitle('顾客分群可视化', y=1.02)
plt.show()

from sklearn.utils import shuffle
import numpy as np
from PIL import Image

# 加载图像
image = Image.open('landscape.jpg')
image = np.array(image) / 255.0  # 归一化

# 获取像素数据
h, w, d = image.shape
pixels = image.reshape(-1, 3)  # 转换为N×3矩阵

# 随机采样加速计算
sample_size = 1000
pixel_sample = shuffle(pixels, random_state=42)[:sample_size]

# 寻找最佳K值（颜色数）
inertias = []
for k in range(1, 11):
    kmeans = KMeans(n_clusters=k, random_state=42)
    kmeans.fit(pixel_sample)
    inertias.append(kmeans.inertia_)

# 根据肘部法则选择K=8
k = 8

# 全像素聚类
kmeans = KMeans(n_clusters=k, random_state=42)
kmeans.fit(pixels)

# 重建压缩图像
compressed_pixels = kmeans.cluster_centers_[kmeans.labels_]
compressed_image = compressed_pixels.reshape(h, w, d)

# 保存结果
Image.fromarray((compressed_image * 255).astype(np.uint8)).save('compressed.jpg')

print(f"原始图像存储: {h*w*24} bits")
print(f"压缩后存储: {h*w*np.log2(k) + k*24} bits")  # 存储标签+颜色表

from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.decomposition import PCA
import matplotlib.pyplot as plt

# 加载文本数据
documents = load_news_corpus()  # 假设返回文本列表

# 文本向量化
vectorizer = TfidfVectorizer(max_features=2000, stop_words='english')
X = vectorizer.fit_transform(documents)

# 降维可视化
pca = PCA(n_components=2)
X_2d = pca.fit_transform(X.toarray())

# 聚类（假设K=6）
kmeans = KMeans(n_clusters=6, random_state=42)
clusters = kmeans.fit_predict(X)

# 可视化聚类结果
plt.figure(figsize=(10, 8))
scatter = plt.scatter(X_2d[:, 0], X_2d[:, 1], c=clusters, cmap='viridis', alpha=0.6)
plt.title('新闻文档聚类可视化')
plt.xlabel('PCA 1')
plt.ylabel('PCA 2')
plt.colorbar(scatter, label='Cluster')
plt.show()

# 查看每个簇的主题词
feature_names = vectorizer.get_feature_names_out()
order_centroids = kmeans.cluster_centers_.argsort()[:, ::-1]

for i in range(6):
    print(f"\nCluster #{i} 主题词:")
    for ind in order_centroids[i, :10]:
        print(f" {feature_names[ind]}", end='')