python-玩转PCA

    PCA降维为计算数据提供了很好的工具，下面，我们以人脸数据集（jaffe）为例，比较前后聚类kmeans采用PCA产生了明显的效果。

     不啰嗦了，上代码：

from PIL import Image
import numpy as np
from numpy import *
def load_image_PIL():
    import os
    from PIL import Image   
    imgs=os.listdir('d:/learning/pcaandknn')
    num=len(imgs)
    dataMat=zeros((num,256*256))
    for i in range(num):
        img=Image.open('d:/learning/pcaandknn/'+imgs[i])     
        img_arr=np.array(np.array(img,dtype='float32').flatten())
        dataMat[i,:]=img_arr
    print(dataMat,shape(dataMat))
    return dataMat
def Modul_pca_front_fit(X):
    from sklearn.cluster import KMeans
    from sklearn.externals import joblib
    clf=KMeans(n_clusters=10)
    clf.fit(X)
#    print(clf.cluster_centers_)
    print(clf.labels_)
    joblib.dump(clf,"d:\\km.pkl")
    clf=joblib.load("d:\\km.pkl")
#    print(clf)
def Modul_pca_after_fit(X):
    from sklearn.cluster import KMeans
    from sklearn.externals import joblib
    from sklearn.decomposition import PCA
    pca=PCA(0.95)
    pca.fit(X)
    X_reduction=pca.transform(X)
    print(shape(X_reduction))
    return X_reduction
#    print(pca.n_components)
#    print(pca.n_components_)   
X=load_image_PIL()
%time Modul_pca_front_fit(X)   #11.1 s
%time X=Modul_pca_after_fit(X) #2.73 s
%time Modul_pca_front_fit(X)  #60 ms

降维后，再执行聚类算法，只需要60ms，速度明显啊。并且维度从65536降到了85维度，是不是很激动啊。