scipy.optimize.linprog: Correct with method=”interior point”, wrong with “simplex”

by

I am using scipy.optimize.linprog. When I use method=”simplex”, I get incorrect answers for larger datasets (it simply gives a result of 0). When I use method=”interior point” for the same datasets and with the same args apart from the method, I get the correct answers.

Why does this happen?

(Unfortunately I am doing an assignment that requires results for both the simplex and interior point methods).

Verified correctness by solving dual problem. Tried setting autoscale to true.

  • For the sake of the assignment, you can try method='revised simplex'. But all of these methods are deprecated, and they will be removed from new releases of SciPy soon, so they should no longer be relied on. Prefer method='highs' for new code. There are separate 'highs-simplex' and 'highs-ipm' methods for simplex and interior-point methods, but they will both give you a basic solution because 'highs-ipm' does crossover at the end by default. As for why it happens: method='simplex' is a very straightforward implementation. It takes a lot more to get a robust LP solver.

    – 

Leave a Comment