dbscan算法实现鸢尾花聚类(python实现)

DBscan算法原理 :

dbscan算法-CSDN博客

法一(调库) :

直接调库 :

import numpy as np
import matplotlib.pyplot as plt
from sklearn import datasets
from sklearn.cluster import DBSCAN
from sklearn.decomposition import PCA
from sklearn.discriminant_analysis import StandardScaler# 加载数据集
iris = datasets.load_iris()
X = iris.data# 数据预处理，标准化数据
scaler = StandardScaler()
X = scaler.fit_transform(X)# 使用DBSCAN聚类算法
dbscan = DBSCAN(eps=0.5, min_samples=5) # 获取DBSCAN聚类对象
y_pred = dbscan.fit_predict(X)# 输出聚类结果
print('聚类结果:', y_pred)# 可视化
pca = PCA(n_components=2)
transformed = pca.fit_transform(X)print(transformed)# 绘制聚类结果
plt.scatter(transformed[:, 0], transformed[:, 1], c=y_pred)
plt.xlabel('Component 1')
plt.ylabel('Component 2')
plt.title('DBSCAN Clustering on Iris Dataset')
plt.show()

其中重要的代码也就两行 :

# 使用DBSCAN聚类算法
dbscan = DBSCAN(eps=0.5, min_samples=5) # 获取DBSCAN聚类对象
y_pred = dbscan.fit_predict(X)

实现效果 :

法二(手写):

思路 : 根据原理实现,可根据具体注释理解(相信一定能够看懂)

import numpy as np
import matplotlib.pyplot as plt
from sklearn import datasets
from sklearn.cluster import DBSCAN
from sklearn.decomposition import PCAdef distance(p1, p2): # 计算两点的欧式距离return np.sqrt(np.sum((p1 - p2) ** 2))def region_query(data, idx , eps): # 获取与data相邻点的下标集合neighbors = [] # 创建空邻居列表for index, point in enumerate(data):if distance(point, data[idx]) <= eps:neighbors.append(index)return neighborsdef expand_cluster(data, labels, point_index, cluster_label, eps, min_samples): # 对点进行扩展neighbors = region_query(data, point_index, eps)if len(neighbors) < min_samples:# 领域内少于min_samples --> 为噪声点labels[point_index] = -1  # 标记为噪声点return Falseelse:labels[point_index] = cluster_label # 标记为当前标签for neighbor_index in neighbors:if labels[neighbor_index] == 0:# 该点未访问过labels[neighbor_index] = cluster_labelexpand_cluster(data, labels, neighbor_index, cluster_label, eps, min_samples) # 继续找下去 , 递归return Truedef dbscan(data, eps, min_samples):n = len(data) # 求数据的长度labels = np.zeros(n)  # 0表示未分类  : 先全部赋值为 0cluster_label = 0 # 簇的数量 / 簇的标记 , 每当一个新的聚类被创建时，cluster_label的值会递增，以便为下一个聚类指定不同的标签。# 类似于BFSfor idx in range(n):# 访问所有点if labels[idx] == 0:# 当前点未访问if expand_cluster(data, labels, idx , cluster_label + 1, eps, min_samples):cluster_label += 1return labels# 准备数据准备
iris = datasets.load_iris()
x = iris.data # 导入鸢尾花数据集# DBSCAN进行聚类
eps = 0.5  # 邻域半径
min_samples = 5  # 最小样本数
labels = dbscan(x, eps, min_samples) # 获取聚类结果print(labels)# 可视化
pca = PCA(n_components=2)
transformed = pca.fit_transform(x)plt.scatter(transformed[:, 0], transformed[:, 1], c=labels)
plt.xlabel('Component 1')
plt.ylabel('Component 2')
plt.title('DBSCAN Clustering on Iris Dataset')
plt.show()

实现效果 :