HW3.3 Projection to the Original Space

"""
Projection to the Original Space
Date:29.06.2019
Author:Yulian Sun
"""
import numpy as np
import matplotlib.pyplot as plt

f= np.loadtxt('../iris.txt', delimiter=',')
file=f[:,0:4]

class PCA:
    def __init__(self,D,N,file):
        self.D = D
        self.N = N
        self.file = file
        self.mu = file.mean(axis=0)
        self.std = file.std(axis=0)

    def data_normalization(self):
        norm_data = np.empty((self.N,self.D))
        norm_data =(self.file - self.mu)/self.std
        return norm_data

    def principle_component(self,data):
        n = len(data)
        cov = 1/n * data.T @ data
        values ,vectors = np.linalg.eig(cov)
        index = np.argsort(-values)# from big to small
        return values ,vectors

    def NRMSE(self,norm_data,vectors,n):
        arr = np.empty_like((norm_data))
        for i in range(4):
            arr[:, i] = vectors[:, i].T @ norm_data.T

        proj = arr[:,0:n] @ vectors[:, 0:n].T
        print(proj)
        error = np.sqrt(np.sum((norm_data - proj) ** 2, axis=0) / len(norm_data))
        return error

N = file.shape[0]
D = file.shape[1]
pca = PCA(D,N,file)
norm_data = pca.data_normalization()
#print(norm_data)
values ,vectors = pca.principle_component(norm_data)
comp_num = 4
error = np.zeros((4,4))
for i in range((1,comp_num)):
    error = pca.NRMSE(norm_data,vectors,i)
    print('file in components {} is :'.format(i+1),error)