Last active
April 27, 2021 17:25
-
-
Save Quwarm/b3624e17fe9a200b24bf32831156ae7a to your computer and use it in GitHub Desktop.
Prekopcsák — Lemire shrinkage
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import numpy as np | |
def prekopcsac_lemire(class_): | |
m, n = class_.shape | |
r = np.corrcoef(class_, rowvar=False) | |
# CALC standardized class_ | |
mean_vec = class_.mean(axis=0) | |
std_vec = class_.std(axis=0, ddof=1) | |
smx = (class_ - np.tile(mean_vec, (m, 1))) / np.tile(std_vec, (m, 1)) | |
# CALC correlation variance | |
w = np.zeros((m, n, n)) | |
for i in range(m): | |
w[i, :, :] = smx[i, 0].transpose() * smx[i, 0] | |
aw = np.zeros((n, n)) | |
for i in range(n): | |
aw[i, :] = w[:, i, :].mean(axis=0) | |
var_r = np.zeros((n, n)) | |
for i in range(n): | |
for j in range(n): | |
var_r[i, j] = (n / (n - 1) ** 3) * sum((w[:, i, j] - aw[i, j]) ** 2) | |
# CALC lambda | |
sum_var_r = sum(sum(var_r)) - sum(np.diag(var_r)) | |
sum_rr = sum(sum(r ** 2)) - sum(np.diag(r) ** 2) | |
lambda_star = sum_var_r / sum_rr | |
lambda_ = max(0, min(1, lambda_star)) | |
# CALC R* | |
multiple = 1 - lambda_ | |
r_star = np.zeros((n, n)) | |
for i in range(n): | |
for j in range(n): | |
r_star[i, j] = r[i, j] * multiple if i != j else 1 | |
# CALC COV* | |
cov_star = np.zeros((n, n)) | |
for i in range(n): | |
for j in range(n): | |
cov_star[i, j] = r_star[i, j] * std_vec[i] * std_vec[j] if i != j else std_vec[i] ** 2 | |
return cov_star, lambda_ | |
def mahalanobis(point_from, point_to, inverse_covariance_matrix): | |
delta = point_from - point_to | |
return max(np.float64(0), np.dot(np.dot(delta, inverse_covariance_matrix), delta)) ** 0.5 | |
def approx(number: float, *, sign, epsilon=1e-4): | |
return number + np.sign(sign) * epsilon | |
test_point = np.array([1., 2.]) | |
class_ = np.array([[1., 1.], [2., 2.]]) | |
T = np.diag(class_.std(axis=0, ddof=1) ** 2) | |
pl_covariance_matrix, lambda_ = prekopcsac_lemire(class_) | |
print("T:", *T) | |
print("COV(*):", *pl_covariance_matrix) | |
print("Lambda:", lambda_) | |
# Первое условие - T является положительно определенной матрицей | |
# (достаточное условие: все собственные значения матрицы T положительны) | |
first_condition = (np.linalg.eig(T)[0] > approx(0., sign=+1)).all() | |
print("All(", np.linalg.eig(T)[0], ") > 0 ? -> ", first_condition, sep='') | |
# Второе условие - лямбда в полуинтервале (0, 1] | |
second_condition = approx(0., sign=+1) < lambda_ <= 1 | |
print("Lambda =", lambda_, "in (0, 1] ? ->", second_condition) | |
# Третье условие - наименьшее собственное значение матрицы COV(*) | |
# должно быть не меньше lambda, умноженной на наименьшее собственное значение T | |
cov_eig = min(np.linalg.eig(pl_covariance_matrix)[0]) | |
lambda_t_eig = lambda_ * min(np.linalg.eig(T)[0]) | |
third_condition = cov_eig >= lambda_t_eig | |
print(cov_eig, ">=", lambda_t_eig, "? ->", third_condition) | |
conditions = [first_condition, second_condition, third_condition] | |
if all(conditions): | |
print("Все три условия выполнены") | |
# Обратная матрица | |
inverse_pl_covariance_matrix = np.linalg.inv(pl_covariance_matrix) | |
# Вычисление расстояний | |
for point_to in [class_.mean(axis=0), *class_]: | |
print("d_M(*) (", test_point, ", ", point_to, ", COV(*)) = ", | |
mahalanobis(test_point, point_to, inverse_pl_covariance_matrix), sep='') | |
else: | |
print("Невыполненные условия (1-3):", *[i for i, x in enumerate(conditions, 1) if not x]) | |
# Вывод: | |
# T: [0.5 0. ] [0. 0.5] | |
# COV(*): [0.5 0.5] [0.5 0.5] | |
# Lambda: 0 | |
# All([0.5 0.5]) > 0 ? -> True | |
# Lambda = 0 in (0, 1] ? -> False | |
# 2.220446049250313e-16 >= 0.0 ? -> True | |
# Невыполненные условия (1-3): 2 | |
# Комментарий: | |
# Лямбда <= 0, условие 2 не выполнено, вычисление результата невозможно |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment