接上文,我們詳細介紹了DBSCANComparison and process with several common clustering algorithms,DBSCANClustering algorithms are the most special,It is a density-based clustering method,聚類前不需要預先指定聚類的個數,接下來將DBSCANAnalysis code sharing
Python代碼：
import numpy as np
import matplotlib.pyplot as plt
from sklearn.datasets import make_blobs
from sklearn.preprocessing import StandardScaler
from sklearn.cluster import DBSCAN
from sklearn import metrics
UNCLASSIFIED = 0
NOISE = -1
# 計算數據點兩兩之間的距離
def getDistanceMatrix(datas):
N, D = np.shape(datas)
dists = np.zeros([N, N])
for i in range(N):
for j in range(N):
vi = datas[i, :]
vj = datas[j, :]
dists[i, j] = np.sqrt(np.dot((vi - vj), (vi - vj)))
return dists
# Look for dotscluster_id 為中心,eps is the radius of all points within the circleid
def find_points_in_eps(point_id, eps, dists):
index = (dists[point_id] <= eps)
return np.where(index == True)[0].tolist()
# Cluster expansion
# dists ： The distance between all data pairs N x N
# labs : Labels for all data labs N,
# cluster_id ： The label of a cluster
# eps ： Density evaluation radius
# seeds： Point to use for cluster expansion
# min_points： Minimum number of points within the radius
def expand_cluster(dists, labs, cluster_id, seeds, eps, min_points):
i = 0
while i < len(seeds):
# Get a proximity point
Pn = seeds[i]
# If the point is marked as NOISE relabel
if labs[Pn] == NOISE:
labs[Pn] = cluster_id
# If the point has not been marked
elif labs[Pn] == UNCLASSIFIED:
# 進行標記,and calculate its neighbors new_seeds
labs[Pn] = cluster_id
new_seeds = find_points_in_eps(Pn, eps, dists)
# 如果 new_seeds long enough to add it toseed 隊列中
if len(new_seeds) >= min_points:
seeds = seeds + new_seeds
i = i + 1
def dbscan(datas, eps, min_points):
# 計算 distance between all points
dists = getDistanceMatrix(datas)
# Initialize the labels of all points as UNCLASSIFIED
n_points = datas.shape[0]
labs = [UNCLASSIFIED] * n_points
cluster_id = 0
# 遍歷所有點
for point_id in range(0, n_points):
# If the current point has already been processed
if not (labs[point_id] == UNCLASSIFIED):
continue
# If it has not been processed, the adjacent points are calculated
seeds = find_points_in_eps(point_id, eps, dists)
# If the number of adjacent points is too small, it will be marked as NOISE
if len(seeds) < min_points:
labs[point_id] = NOISE
else:
# Otherwise, a round of cluster expansion is started
cluster_id = cluster_id + 1
# 標記當前點
labs[point_id] = cluster_id
expand_cluster(dists, labs, cluster_id, seeds, eps, min_points)
return labs, cluster_id
# 繪圖
def draw_cluster(datas, labs, n_cluster):
plt.cla()
colors = [plt.cm.Spectral(each)
for each in np.linspace(0, 1, n_cluster)]
for i, lab in enumerate(labs):
if lab == NOISE:
plt.scatter(datas[i, 0], datas[i, 0], s=16., color=(0, 0, 0))
else:
plt.scatter(datas[i, 0], datas[i, 0], s=16., color=colors[lab - 1])
plt.show()
if __name__ == "__main__":
## 數據1
# centers = [[1, 1], [-1, -1], [1, -1]]
# datas, labels_true = make_blobs(n_samples=750, centers=centers, cluster_std=0.4,
# random_state=0)
## 數據2
file_name = "spiral"
with open(file_name + ".txt", "r", encoding="utf-8") as f:#
lines = f.read().splitlines()
lines = [line.split("\t")[:-1] for line in lines]
datas = np.array(lines).astype(np.float32)
# 數據正則化
datas = StandardScaler().fit_transform(datas)
eps = 20#半徑
min_points = 0
labs, cluster_id = dbscan(datas, eps=eps, min_points=min_points)
print("labs of my dbscan")
print(labs)
db = DBSCAN(eps=eps, min_samples=min_points).fit(datas)
skl_labels = db.labels_
print("labs of sk-DBSCAN")
print(skl_labels)
draw_cluster(datas, labs, cluster_id)
MATLAB代碼如下：
data=xlsread('C:/Users/zhichu/Desktop/附件1 弱覆蓋柵格數據(篩選).csv');%導入數據
x=data(:,1);
y=data(:,2);
figure('Name','散點圖分布','NumberTitle','off');
scatter(x,y,0.5,'k')
axis([0,2499,0,2499])
epsilon=20;%Maximum clustering distance between base stations,自己根據需要設置
minpts=1;%Minimum number of clusters
idx=dbscan([x,y],epsilon,minpts);
length(unique(idx))
[gc,grps]=groupcounts(idx)
sortrows([gc,grps],'descend')
figure('Name','DBSCAN聚類結果','NumberTitle','off');
gscatter(x,y,idx,[],[],1,'doleg','off')
xlabel('x坐標');ylabel('y坐標')
This function can also calculate the clustering results when faced with hundreds of thousands of data,Therefore, everyone is faced with large dataDBSCANThis built-in function can be used for clustering problems,前提是你的MATLAB版本要高於2019And the Statistics and Machine Learning Toolbox is installed（ Statistics and Machine LearningToolbox）.
