1

I am trying the following code with Python 2.7.12 and Numpy 1.11.0. 

import numpy as np
from scipy.optimize import linprog

A = np.matrix([[1,0,0,1], [0, 1, 0 ,1], [0, 1, 1, 0]])
c = np.zeros(A.shape[1])
res = linprog(c,A_eq = A, b_eq = [0,1,-1], options=dict(bland=True))
print (res.x)
print (A.dot([0, 1, -2, 0]))

The output of the above is

nan
[[ 0  1 -1]]

So scipy.optimize.linprog does not find a solution even when one exists as evident by output of the dot multiplication of A_eq with [0, 1, -2, 0].

A similar question was asked here and I tried the solutions suggested there (i.e. adding options=dict(bland=True) or updating the tolerance value). I still get the same erroneous output as posted above. What could be the reason for this behaviour? Thank you.

Hummels420
  • 21
  • 3
  • Possible duplicate of [scipy.optimize.linprog unable to find a feasible starting point despite a feasible answer clearly exists](https://stackoverflow.com/questions/29941958/scipy-optimize-linprog-unable-to-find-a-feasible-starting-point-despite-a-feasib) – Jonas Adler Jul 31 '17 at 12:39
  • Hi Jonas, Thank you for your comment. Although I had looked at that question and tried the solutions suggested there (i.e. adding options=dict(bland=True) or updating the tolerance value). I still get the same erroneous output as posted in the question. – Hummels420 Jul 31 '17 at 12:47

1 Answers1

1

I am the OP and the solution was adding the bounds explicitly as follows:

 res = linprog(c, A_eq = A, b_eq = [0,1,-1], bounds=(None, None))

I was under the impression that by default, linprog assumes no bounds on the solution but in fact the default bounds are (0, None) i.e. non-negative values.

This is mentioned in the docs:

bounds : sequence, optional

(min, max) pairs for each element in x, defining the bounds on that parameter. Use None for one of min or max when there is no bound in that direction. By default bounds are (0, None) (non-negative) ...

Community
  • 1
  • 1
Hummels420
  • 21
  • 3