17.HDBSCAN Code

[목적]

• 비지도학습 중 하나인 Clustering 중 HDBSCAN 실습
• For loop 활용 Hyperparameter(1개) 변경시켜 가며 실습 진행

[Process]

1. Define X’s
2. Modeling

  !pip install hdbscan

  import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
import sklearn.datasets as data
import hdbscan

  #plt 와 sns Setting
%matplotlib inline
sns.set_context('poster')
sns.set_style('white')
sns.set_color_codes()
plot_kwds = {'alpha' : 0.5, 's' : 80, 'linewidths':0}
plt.rcParams["figure.figsize"] = [9,7]

  # Sample Data 만들기
num=100
moons, _ = data.make_moons(n_samples=num, noise=0.01)
blobs, _ = data.make_blobs(n_samples=num, centers=[(-0.75,2.25), (1.0, -2.0)], cluster_std=0.25)
blobs2, _ = data.make_blobs(n_samples=num, centers=[(2,2.25), (-1, -2.0)], cluster_std=0.4)
test_data = np.vstack([moons, blobs,blobs2])
plt.scatter(test_data.T[0], test_data.T[1], color='b', **plot_kwds)
plt.show()

[HDBSCAN]

• Hyperparameter Tuning using for Loop [HDBSCAN Parameters]
• Packge : https://hdbscan.readthedocs.io/en/latest/how_hdbscan_works.html
• min_cluster : Cluster 안에 적어도 몇 개가 있어야 하는지
• cluster_selection_epsilon : combining HDBSACN with DBSCAN

  minsize = [3, 5, 10, 15, 20, 30]
for m in minsize:
    print("min_cluster_size : {}".format(m))
    db = hdbscan.HDBSCAN(min_cluster_size=m).fit(test_data)
    palette = sns.color_palette()
    cluster_colors = [palette[col]
                    if col >= 0 else (0.5, 0.5, 0.5) for col in
                    db.labels_]
    plt.scatter(test_data.T[0], test_data.T[1], c=cluster_colors, **plot_kwds)
    plt.show()

min_cluster_size : 20 min_cluster_size : 30 min_cluster_size : 50 min_cluster_size : 70

• parameter 하나만 조절해도 되어서 더 간단하다!