95 lines
3.5 KiB
Python
95 lines
3.5 KiB
Python
|
from __future__ import annotations
|
||
|
|
|
||
|
|
from math import floor as mfloor
|
||
|
|
|
||
|
|
from sympy.polys.domains import ZZ, QQ
|
||
|
|
from sympy.polys.matrices.exceptions import DMRankError, DMShapeError, DMValueError, DMDomainError
|
||
|
|
|
||
|
|
|
||
|
|
def _ddm_lll(x, delta=QQ(3, 4), return_transform=False):
|
||
|
|
if QQ(1, 4) >= delta or delta >= QQ(1, 1):
|
||
|
|
raise DMValueError("delta must lie in range (0.25, 1)")
|
||
|
|
if x.shape[0] > x.shape[1]:
|
||
|
|
raise DMShapeError("input matrix must have shape (m, n) with m <= n")
|
||
|
|
if x.domain != ZZ:
|
||
|
|
raise DMDomainError("input matrix domain must be ZZ")
|
||
|
|
m = x.shape[0]
|
||
|
|
n = x.shape[1]
|
||
|
|
k = 1
|
||
|
|
y = x.copy()
|
||
|
|
y_star = x.zeros((m, n), QQ)
|
||
|
|
mu = x.zeros((m, m), QQ)
|
||
|
|
g_star = [QQ(0, 1) for _ in range(m)]
|
||
|
|
half = QQ(1, 2)
|
||
|
|
T = x.eye(m, ZZ) if return_transform else None
|
||
|
|
linear_dependent_error = "input matrix contains linearly dependent rows"
|
||
|
|
|
||
|
|
def closest_integer(x):
|
||
|
|
return ZZ(mfloor(x + half))
|
||
|
|
|
||
|
|
def lovasz_condition(k: int) -> bool:
|
||
|
|
return g_star[k] >= ((delta - mu[k][k - 1] ** 2) * g_star[k - 1])
|
||
|
|
|
||
|
|
def mu_small(k: int, j: int) -> bool:
|
||
|
|
return abs(mu[k][j]) <= half
|
||
|
|
|
||
|
|
def dot_rows(x, y, rows: tuple[int, int]):
|
||
|
|
return sum([x[rows[0]][z] * y[rows[1]][z] for z in range(x.shape[1])])
|
||
|
|
|
||
|
|
def reduce_row(T, mu, y, rows: tuple[int, int]):
|
||
|
|
r = closest_integer(mu[rows[0]][rows[1]])
|
||
|
|
y[rows[0]] = [y[rows[0]][z] - r * y[rows[1]][z] for z in range(n)]
|
||
|
|
mu[rows[0]][:rows[1]] = [mu[rows[0]][z] - r * mu[rows[1]][z] for z in range(rows[1])]
|
||
|
|
mu[rows[0]][rows[1]] -= r
|
||
|
|
if return_transform:
|
||
|
|
T[rows[0]] = [T[rows[0]][z] - r * T[rows[1]][z] for z in range(m)]
|
||
|
|
|
||
|
|
for i in range(m):
|
||
|
|
y_star[i] = [QQ.convert_from(z, ZZ) for z in y[i]]
|
||
|
|
for j in range(i):
|
||
|
|
row_dot = dot_rows(y, y_star, (i, j))
|
||
|
|
try:
|
||
|
|
mu[i][j] = row_dot / g_star[j]
|
||
|
|
except ZeroDivisionError:
|
||
|
|
raise DMRankError(linear_dependent_error)
|
||
|
|
y_star[i] = [y_star[i][z] - mu[i][j] * y_star[j][z] for z in range(n)]
|
||
|
|
g_star[i] = dot_rows(y_star, y_star, (i, i))
|
||
|
|
while k < m:
|
||
|
|
if not mu_small(k, k - 1):
|
||
|
|
reduce_row(T, mu, y, (k, k - 1))
|
||
|
|
if lovasz_condition(k):
|
||
|
|
for l in range(k - 2, -1, -1):
|
||
|
|
if not mu_small(k, l):
|
||
|
|
reduce_row(T, mu, y, (k, l))
|
||
|
|
k += 1
|
||
|
|
else:
|
||
|
|
nu = mu[k][k - 1]
|
||
|
|
alpha = g_star[k] + nu ** 2 * g_star[k - 1]
|
||
|
|
try:
|
||
|
|
beta = g_star[k - 1] / alpha
|
||
|
|
except ZeroDivisionError:
|
||
|
|
raise DMRankError(linear_dependent_error)
|
||
|
|
mu[k][k - 1] = nu * beta
|
||
|
|
g_star[k] = g_star[k] * beta
|
||
|
|
g_star[k - 1] = alpha
|
||
|
|
y[k], y[k - 1] = y[k - 1], y[k]
|
||
|
|
mu[k][:k - 1], mu[k - 1][:k - 1] = mu[k - 1][:k - 1], mu[k][:k - 1]
|
||
|
|
for i in range(k + 1, m):
|
||
|
|
xi = mu[i][k]
|
||
|
|
mu[i][k] = mu[i][k - 1] - nu * xi
|
||
|
|
mu[i][k - 1] = mu[k][k - 1] * mu[i][k] + xi
|
||
|
|
if return_transform:
|
||
|
|
T[k], T[k - 1] = T[k - 1], T[k]
|
||
|
|
k = max(k - 1, 1)
|
||
|
|
assert all([lovasz_condition(i) for i in range(1, m)])
|
||
|
|
assert all([mu_small(i, j) for i in range(m) for j in range(i)])
|
||
|
|
return y, T
|
||
|
|
|
||
|
|
|
||
|
|
def ddm_lll(x, delta=QQ(3, 4)):
|
||
|
|
return _ddm_lll(x, delta=delta, return_transform=False)[0]
|
||
|
|
|
||
|
|
|
||
|
|
def ddm_lll_transform(x, delta=QQ(3, 4)):
|
||
|
|
return _ddm_lll(x, delta=delta, return_transform=True)
|