【python】numpy实现PCA降维

import numpy as np

class PCA:
    def __init__(self, fileName, splitBy=' '):
        self.readData = np.array([line.split(splitBy) for line in open(fileName).readlines()], dtype='float')   # 读取数据为numpy矩阵
        nan_index = np.where(np.isnan(self.readData))   # 获取Nan值的坐标
        self.readData[nan_index] = np.take(np.nanmean(self.readData, axis=0), nan_index[1])     # 将Nan值替换为列均值
        self.initData = self.readData - self.readData.mean(axis=0)  # 得到去均值的数据矩阵
        self.Cov = np.cov(self.initData.T)  # 计算其协方差矩阵
        self.EValue, self.EVector = np.linalg.eig(self.Cov)   # 计算其特征值和特征向量

    def RecudeDimension(self, dim):     # 计算降维后的数据,前dim总方差贡献率，前dim各方差贡献率
        return np.dot(self.initData, self.EVector[:, :dim]), \
               [self.EValue[i]/sum(self.EValue) for i in range(dim)], \
               sum(self.EValue[:dim])/sum(self.EValue)

if __name__ == '__main__':
    pca = PCA('./PCA/secom.data')
    data, vals, vals_sum = pca.RecudeDimension(20)
    print('降维后的数据:', data)
    print('前20个特征的方差贡献率：', vals)
    print('前20个特征的总方差贡献率：', vals_sum)

数据集链接：(https://pan.baidu.com/s/1BB3TRVvK-BALs_Uwe0ejIw
提取码：tdlp