게으른 우루루

agglomerative algorithm군집 알고리즘을 적용하는데 어려운 점은 

algorithm이 잘 작동하는지 평가하거나 여러 algorithm의 출력을 비교하기가 매우 어려움

# library import

import matplotlib

import matplotlib.pyplot as plt

import numpy as np

import mglearn

from sklearn.datasets import make_moons

from sklearn.preprocessing import StandardScaler

from sklearn.cluster import KMeans, AgglomerativeClustering, DBSCAN

from sklearn.metrics.cluster import adjusted_rand_score

# matplotlib 설정

matplotlib.rc('font', family='AppleGothic') # 한글출력

plt.rcParams['axes.unicode_minus'] = False # 축 -표시

# dataset

x, y = make_moons(n_samples=200, noise=0.05, random_state=0)

# pre-processing

scaler = StandardScaler().fit(x)

x_scaled = scaler.transform(x)

# model list

algorithms = [KMeans(n_clusters=2),

              AgglomerativeClustering(n_clusters=2),

                      DBSCAN()]

kmean_pred = algorithms[0].fit_predict(x_scaled)

# random cluster

random_state = np.random.RandomState(seed=0)

random_clusters = random_state.randint(low=0, high=2, size=len(x))

### visualization

fig, axes = plt.subplots(2, 2,

                         subplot_kw={'xticks':(), 'yticks':()})

axes[0,0].scatter(x_scaled[:, 0], x_scaled[:, 1], c=random_clusters,

                cmap=mglearn.cm3, s=60, edgecolors='k')

axes[0,0].set_title('random assign - ARI: {:.3f}'.format(

    adjusted_rand_score(y, random_clusters))) # 실제, 모델로 구한 클러스터

axes[0,1].scatter(x_scaled[:, 0], x_scaled[:, 1], c=kmean_pred,

                  cmap=mglearn.cm3, s=60, edgecolors='k')

axes[0,1].set_title('{} - ARI: {:.3f}'.format(algorithms[0].__class__.__name__, #__class__.__name__ ==> 클래스에서 이름 속성

                                              adjusted_rand_score(y, kmean_pred)))

for ax, algorithm in zip(axes.ravel()[2:], algorithms[1:]):

    clusters = algorithm.fit_predict(x_scaled)

    ax.scatter(x_scaled[:, 0], x_scaled[:, 1], c=clusters,

               cmap=mglearn.cm3, s=60, edgecolors='k')

    ax.set_title('{} - ARI: {:.3f}'.format(algorithm.__class__.__name__,

                                           adjusted_rand_score(y, clusters)))

plt.show()

# library import

from sklearn.metrics.cluster import silhouette_score

# datasets

x, y = make_moons(n_samples=300, noise=0.05, random_state=0)

# pre-processing

scaler = StandardScaler().fit(x)

x_scaled = scaler.transform(x)

# random cluster

random_state = np.random.RandomState(seed=0)

random_clusters = random_state.randint(0, 2, size=len(x)) # randint(low, high, size)

# K-Means 모델

kmeans = KMeans(n_clusters=2)

kmeans_pred = kmeans.fit_predict(x_scaled)

### visualization

fig, axes = plt.subplots(2, 2,

                         subplot_kw={'xticks':(), 'yticks':()})

ax = axes.ravel()

# random cluster 그리기

ax[0].scatter(x_scaled[:, 0], x_scaled[:,1], c=random_clusters,

            cmap=mglearn.cm2, edgecolors='k', s=60)

ax[0].set_title('random assign - silhouette score: {:.3f}'.format(silhouette_score(x_scaled, random_clusters))) # silhouette_score(훈련데이터, 예측한 클러스터)

# K-Means 그리기

ax[1].scatter(x_scaled[:, 0], x_scaled[:, 1], c=kmeans_pred,

           cmap=mglearn.cm2, edgecolors='k', s=60)

ax[1].set_title('{} - silhouette score: {:.3f}'.format(kmeans.__class__.__name__,

                                                 silhouette_score(x_scaled, kmeans_pred)))

# AgglomerativeClustering과 DBSCAN 그리기

algorithms = [AgglomerativeClustering(n_clusters=2), DBSCAN()]

for ax, model in zip(ax[2:], algorithms):

    cluster = model.fit_predict(x_scaled)

    ax.scatter(x_scaled[:, 0], x_scaled[:, 1], c=cluster,

               cmap=mglearn.cm2, edgecolors='k', s=60)

    ax.set_title('{} - silhouette score: {:.3f}'.format(model.__class__.__name__,

                                                  silhouette_score(x_scaled, cluster)))

plt.show()