近似因子模型

import numpy as np
from numpy.random import multivariate_normal, random, poisson
from numpy.linalg import eigh, norm, svd
from scipy.stats import expon
import matplotlib.pyplot as plt
from sklearn import linear_model

from functools import partial

def generate_sample(num_of_sample, dim_of_sample):
 return multivariate_normal(np.zeros([dim_of_sample]), np.eye(dim_of_sample), num_of_sample)

def generate_E(left_lambda, right_lambda, samples):
 def generate_matrix(dim, _lambda):

  base_matrix = np.zeros([dim, dim])
  for i in range(base_matrix.shape[0]):
   for j in range(i):
    base_matrix[i][j] = random()

  for i in range(base_matrix.shape[0]):
   for j in range(base_matrix.shape[0])[i:]:
    base_matrix[i][j] = base_matrix[j][i]

  orth_matrix = eigh(base_matrix)[1]
  lambda_vector = np.random.random_sample((base_matrix.shape[0] ,)) * _lambda

  return np.dot(np.dot(orth_matrix.T, np.diag(lambda_vector)), orth_matrix)

 return np.dot(np.dot(generate_matrix(samples.shape[0], left_lambda), samples), generate_matrix(samples.shape[1], right_lambda))


def E_for_singular(max_lambda, lambda_num, num_of_sample, dim_of_sample):
 ListRequire = []
 lambda_vector = np.linspace(max_lambda / lambda_num, max_lambda, lambda_num - 1)

 for _lambda in lambda_vector:
  singular_vector = svd(generate_E(_lambda, _lambda, generate_sample(num_of_sample, dim_of_sample)))[1]
  ListRequire.append(singular_vector[-1] / singular_vector[0])

 return (lambda_vector ,ListRequire)

'''
E_vectors = E_for_singular(10, 100, 1000, 10)
print E_vectors[0]
print E_vectors[1]
'''
#The conclusion of E_for_singula