猫史档案馆

自己写的一个pca框架，可以跑一下加深印象：

import numpy as np
from matplotlib import pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
import warnings
class PCA():
    def __init__(self):
        self.data = None
        self.dimension = 1
        self.decline = None
        self.value = None
    def input(self,data,dimension):
        data = np.array(data).T
        if data.shape[0] != dimension:
            raise Exception("Check your data's dimension,they didn't fit.")
        else:
            self.data = data
            mean = np.array([np.mean(self.data,axis=1)]).T
            self.data = self.data - np.tile(mean,(1,data.shape[1]))
    def set_dimension(self,x):
        if x < 1 or int(x) != x:
            raise Exception("Dimension can't be this:(Int and larger than 0)", x)
        else:
            self.dimension = x
    def Decline(self):
        cov = np.cov(self.data)
        self.value, vector = np.linalg.eig(cov)
        self.decline = vector[0:self.dimension].dot(self.data)
        self.decline = (self.decline.T).tolist()
        result = []
        for i in range(len(self.decline)):
            result.append(self.decline[i])
        return result[::-1]
    def check(self,plot=False):
        if plot == True:
            plt.plot(self.value)
            plt.show()
        return np.sum(self.value)


if __name__ == '__main__':
    data1 = [[1,2],[2,4],[3,6],[4,8],[5,10]]
    pca = PCA()
    pca.input(data1,2)
    pca.set_dimension(1)
    result = pca.Decline()

    pca.check(plot=True)