主成分分析,Principal Component Analysis
计算协方差矩阵,由于是hermitian的,所以100%可正交对角化而且特征值一定非负
求出特征值特征向量以后,取特征值比较大的那几个方向构成线性空间,把数据投影上去就OK了
补详细公式推导:
http://blog.csdn.net/u013648367/article/details/73824049
import os
import numpy as np
import matplotlib.pylab as plt
def dataLoader(file):
return np.array([ i.split() for i in open(file)],dtype="float")
def pca(dataSet,rank=0):
means = np.mean(dataSet,axis=0)
dataSet = dataSet - means #去中心化 ,使得E(X),E(Y),... =0, 方便求方差
#plt.scatter(*dataSet.T)
lenth = dataSet.shape[0]
if rank == 0:
rank = dataSet.shape[1]+1
dataSet = np.matrix(dataSet)
covMat = dataSet.T*dataSet/(lenth-1)
eigenVals, eigenVec = np.linalg.eigh(covMat)
pick_id = eigenVals.argsort()[::-1][:rank] #从小到大排序,返回数组序号,然后倒着取,取前r个
eigenVec_picked = eigenVec[:,pick_id]#取出排序后的特征向量
#print(