Is there a way to prevent "crossover" iterations when using scipy.optimize.libprog's solver "highs-ipm"?

Question

Is there an option to prevent "crossover" when calling highs-ipm solver via scipy.optimize.linprog? It appears that the "standalone" solver (see https://github.com/ERGO-Code/HiGHS) has an option "--run_crossover arg". I just wonder if this is reachable via Python's scipy interface?

The documentation page for linpropg does not list this as a parameter, which of course suggests the negative answer. https://docs.scipy.org/doc/scipy/reference/optimize.linprog-highs-ipm.html

Answer 1

Since version 1.11.1 of scipy, it is possible to pass unknown options to the solver like run_crossover=False. Scipy will generate a warning but the options will be passed to the solver.

It is possible to check the number of iterations performed in the crossover with sol.crossover_nit. The result should be 0.

from scipy.optimize import linprog

# Define c, A_ub, b_ub, A_eq, b_eq, bounds
...

# Define options
solver = 'highs-ipm'
params = {'run_crossover': False}

# Optimize
sol = linprog(c, A_ub, b_ub, A_eq, b_eq, bounds, method=solver, options=params)

# Check the number of iteration in crossover
print(sol.crossover_nit)  # it should be 0

You can check all options HighsOptions.pxd