Clustering of machine learning - the formation of DBSCAN deductive organization

​
​

Event address: CSDN 21-day Learning Challenge

Introduction

There is a vassal called DBSCAN clustering, which is a member of the Density Clustering Alliance.Everyone here has a consensus that the organization is king. At the beginning, a person was an organization, and they didn’t know each other. Later, some people had some friends around them, and they found that they could get some different information from others.And this information sometimes brings a business opportunity, so they get to know friends of friends, and later, they call friends, friends of friends, friends of friends of friends...all together, each otherA merger creates an organization, and those who enter it cannot enter other organizations.As a result, many organizations have formed in this country, and there is no need for the number of organizations designated by the state. These organizations are formed spontaneously according to the relationship between distance and nearness.For members within the organization, when they have made enough friends, they can become core members.There are strong and weak organizations, but they will not annex each other, which can make the country have a very stable political situation.Of course, when the number of people in the organization is too small, it is equivalent to noise and has no right to speak.
This is the national conditions of this country, is it a little similar to our world!The above attempts to explain the principle of DBSCAN in a popular expression, the following is a rigorous and brief description, come on, on the concept!

The principle of DBSCAN

The basic concepts are as follows:

The algorithm flow is as follows:

There are many articles describing the principle of DBSCAN in detail. Here is just a brief summary and personal understanding. Readers who are not addicted can study it in depth.The case below!

Case: Clustering of Special Shape Area Data

The code is as follows:

import numpy as npimport matplotlib.pyplot as pltimport matplotlib.colorsfrom sklearn.cluster import DBSCANif __name__ == "__main__":# datat = np.arange(0, 2*np.pi, 0.1)data1 = np.vstack((np.cos(t), np.sin(t))).Tdata2 = np.vstack((2*np.cos(t), 2*np.sin(t))).Tdata3 = np.vstack((3*np.cos(t), 3*np.sin(t))).Tdata = np.vstack((data1, data2, data3))# Parameters: (field radius, the minimum number of neighbors of the core object in the field radius), here prepare multiple sets of parameters for comparisonparams = ((0.5, 3), (0.5, 5), (0.5, 10), (1., 3))# drawing preparationmatplotlib.rcParams['font.sans-serif'] = ['SimHei']matplotlib.rcParams['axes.unicode_minus'] = Falseplt.figure(figsize=(9, 7), facecolor='w')plt.suptitle('DBSCAN clustering application', fontsize=15)# loop through all parametersfor i in range(4):eps, min_samples = params[i]# Build the model and cluster the datamodel = DBSCAN(eps=eps, min_samples=min_samples)model.fit(data)y_hat = model.labels_# count the number of clustersy_unique = np.unique(y_hat)n_clusters = y_unique.sizeprint('The number of clusters is:', n_clusters)# draw clustering resultsplt.subplot(2, 2, i+1)plt.scatter(data[:, 0], data[:, 1], s=40, c=y_hat, cmap=plt.cm.Spectral, edgecolors="k")plt.grid(visible=True, ls=":", color="#808080")plt.title(r'$\epsilon$ = %.1f m = %d, number of clusters: %d' % (eps, min_samples, n_clusters), fontsize=12)plt.tight_layout()plt.subplots_adjust(top=0.9)# plt.show()plt.savefig('dbscan.png', dpi=800)

The results are as follows:

The author's level is limited, please leave a message to correct me if there is any inadequacy

​