Question

我想解决一个矩形系统（在解决方案中有任意参数）。如果没有，我想在矩阵中添加行，直到它为正方形。

print matrix_a

print vector_b

print len(matrix_a),len(matrix_a[0])

给出：

 [[0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0],
 [0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0],
 [0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0],
 [0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1],
 [0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0],
 [1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0],
 [1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
 [0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
 [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
 [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0],
 [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1]]

[2, 0, 2, 1, 2, 1, 1, 1, 1, 1, 1]

11 26

我的完整代码位于http://codepad.org/tSgstpYe

如你所见，我有一个系统Ax = b。我知道每个x值x1，x2 ..必须是1或0，我希望有这个限制，系统应该只有一个解决方案。

实际上我期望的答案是x = [0,1,0,0,0,0,0,0,0,1,0,0,0,1,0,0,0,1,0 ，0,0,1,0,0,0,0]

我看着 http://docs.scipy.org/doc/numpy/reference/generated/numpy.linalg.solve.html#numpy.linalg.solve 但它似乎只采用方形矩阵。

任何有关解决此系统的帮助都会很棒！

Answer 1

这是一个简单的实现（具有硬编码阈值），但它提供了您正在寻找的测试数据解决方案。

它基于Iteratively Reweighted Least Squares。

from numpy import abs, diag, dot, zeros
from numpy.linalg import lstsq, inv, norm

def irls(A, b, p= 1):
    """Solve least squares problem min ||x||_p s.t. Ax= b."""
    x, p, e= zeros((A.shape[1], 1)), p/ 2.- 1, 1.
    xp= lstsq(A, b)[0]
    for k in xrange(1000):
        if e< 1e-6: break
        Q= dot(diag(1./ (xp** 2+ e)** p), A.T)
        x= dot(dot(Q, inv(dot(A, Q))), b)
        x[abs(x)< 1e-1]= 0
        if norm(x- xp)< 1e-2* e** .5: e*= 1e-1
        xp= x
    return k, x.round()

Answer 2

根据您期望的输入，使用简单的树搜索算法可能会更好。您的结果向量包含相对较低的数字，允许尽早切断大多数树枝。我尝试实现此算法会在0.2秒后产生预期结果：

def testSolution(a, b, x):
  result = 0
  for i in range(len(b)):
    n = 0
    for j in range(len(a[i])):
      n += a[i][j] * x[j]
    if n < b[i]:
      result = -1
    elif n > b[i]:
      return 1
  return result

def solve(a, b):
  def solveStep(a, b, result, step):
    if step >= len(result):
      return False

    result[step] = 1
    areWeThere = testSolution(a, b, result)
    if areWeThere == 0:
      return True
    elif areWeThere < 0 and solveStep(a, b, result, step + 1):
      return True
    result[step] = 0
    return solveStep(a, b, result, step + 1)

  result = map(lambda x: 0, range(len(a[0])))
  if solveStep(a, b, result, 0):
    return result
  else:
    return None

matrix_a = [[0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0],
 [0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0],
 [0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0],
 [0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1],
 [0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0],
 [1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0],
 [1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
 [0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
 [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
 [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0],
 [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1]]

vector_b = [2, 0, 2, 1, 2, 1, 1, 1, 1, 1, 1]

print solve(matrix_a, vector_b)

这必须测试您输入的1325个可能的向量，这比所有可能的结果（6700万）少得多。最糟糕的情况当然是6700万次测试。

Answer 3

让Ax = b成为系统，然后A|b成为增强矩阵

有三种可能性

没有解决方案iff rank(A) < rank(A|b)
iff rank(A) = rank(A|b) = n
iff rank(A) = rank(A|b) < n

其中n是未知数。

解决python中的矩形矩阵，得到任意参数的解

3 个答案: