게으른 우루루

0. 살펴보기

1. PCA, NMF, k-평균에서 추출한 성분을 바탕으로 시각화

# library import

from sklearn.datasets import fetch_lfw_people

from sklearn.model_selection import train_test_split

import matplotlib.pyplot as plt

import matplotlib

import numpy as np

from sklearn.decomposition import NMF

from sklearn.decomposition import PCA

from sklearn.cluster import KMeans

# matplotlib 설정

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

plt.rcParams['axes.unicode_minus'] = False

# people객체 생성

people = fetch_lfw_people(min_faces_per_person=20, resize=0.7, color=False) # 겹치치 않는 최소사람 수, 비율, 흑백

image_shape = people.images[0].shape 

# 특성 50개 추출(True)

idx = np.zeros(people.target.shape, dtype=np.bool) # 3023개의 False생성

for target in np.unique(people.target): 

    idx[np.where(people.target == target)[0][:50]] = 1 

# 데이터 추출

x_people = people.data[idx]

y_people = people.target[idx]

# 데이터 분할

x_train, x_test, y_train, y_test = train_test_split(

    x_people, y_people, # 분할할 데이터

    stratify=y_people, random_state=0) # 그룹, 랜덤상태

# 모델생성 및 학습

nmf = NMF(n_components=100, random_state=0).fit(x_train) 

pca = PCA(n_components=100, random_state=0).fit(x_train)

kmeans = KMeans(n_clusters=100, random_state=0).fit(x_train)

### pca.transform(x_test): 주성분 갯수만큼 차원을 줄인 데이터

### pca.inverse_transform(pca.transform(x_test)): 차원을 줄인 주성분으로 원래 데이터에 매핑

### kmeans.cluster_centers_[kmeans.predict(x_test)] k-평균 중심에서 k-평균으로 구한 라벨을 행 인덱스로 추출하여 재구성

x_reconst_nmf = np.dot(nmf.transform(x_test), nmf.components_) # nmf.inverse_transform(nmf.transform(x_test)) 와 동일

x_reconst_pca = pca.inverse_transform(pca.transform(x_test))

x_reconst_kmeans = kmeans.cluster_centers_[kmeans.predict(x_test)]

# visualization

fig, axes = plt.subplots(3, 5,

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

for ax, comp_nmf, comp_pca, comp_kmeans in zip(

        axes.T, nmf.components_, pca.components_, kmeans.cluster_centers_): # 행방향 순서로 그림을 채우기 때문에 Transfrom을 

    ax[0].imshow(comp_nmf.reshape(image_shape))

    ax[1].imshow(comp_pca.reshape(image_shape))

    ax[2].imshow(comp_kmeans.reshape(image_shape))

axes[0, 0].set_ylabel('nmf')

axes[1, 0].set_ylabel('pca')

axes[2, 0].set_ylabel('kmeans')

plt.gray()

plt.show()

# library import 

fig, axes = plt.subplots(4, 5,

                         subplot_kw={'xticks':(), 'yticks':()}) # subplot 축 없애기

for ax, orig, rec_nmf, rec_pca, rec_kmeans in zip(

        axes.T, x_test, x_reconst_nmf, x_reconst_pca, x_reconst_kmeans):

    ax[0].imshow(orig.reshape(image_shape))

    ax[1].imshow(rec_nmf.reshape(image_shape))

    ax[2].imshow(rec_pca.reshape(image_shape))

    ax[3].imshow(rec_kmeans.reshape(image_shape))

for i, name in zip(range(4), ['원본', 'kmeans', 'pca', 'nmf']):

    axes[i, 0].set_ylabel(name)

plt.gray()

plt.show()

2. 특성의 수보다 많은 클러스터중심을 사용한 k-평균

# library import

from sklearn.datasets import make_moons

# dataset생성

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

# 모델 생성 및 학습, 예측

kmeans = KMeans(n_clusters=10, random_state=0).fit(x)

y_pred = kmeans.predict(x)

plt.scatter(x[:, 0], x[:, 1], c=y_pred, s=60, cmap='Paired', edgecolors='black', alpha=0.7) # cmap: palette

plt.scatter(kmeans.cluster_centers_[:, 0], kmeans.cluster_centers_[:, 1], c=np.unique(kmeans.labels_), # c: group

            s=100, marker='s', linewidths=2, cmap='Paired', edgecolors='black')

plt.xlabel('특성 0')

plt.ylabel('특성 1')

plt.show()