BUG: SIMPLEX_STRATEGY_PRIMAL strategy fails

[mckib2]

The following problem is solved correctly by all other simplex strategies except SIMPLEX_STRATEGY_PRIMAL:

import numpy as np
np.random.seed(0)
m_eq, m_ub, n = 10, 20, 50
c = np.random.rand(n)-0.5
A_ub = np.random.rand(m_ub, n)-0.5
b_ub = np.random.rand(m_ub)-0.5
A_eq = np.random.rand(m_eq, n)-0.5
b_eq = np.random.rand(m_eq)-0.5
lb = -np.random.rand(n)
ub = np.random.rand(n)
lb[lb<-np.random.rand()] = -np.inf
ub[ub>np.random.rand()] = np.inf

or the following MPS file:
simplex_primal_test.txt

The following options were used:

solver = simplex
simplex_strategy = 4
presolve = on # or off, both fail

The HiGHS call was run as follows:

./highs --options_file=opt simplex_primal_test.mps

The following simplex strategies were able to achieve optimal solution:

  SIMPLEX_STRATEGY_MIN,
  SIMPLEX_STRATEGY_DUAL,
  SIMPLEX_STRATEGY_DUAL_TASKS,
  SIMPLEX_STRATEGY_DUAL_MULTI,

I will also note that the simplex_strategy SIMPLEX_STRATEGY_NUM raises the following warning:

WARNING: checkOptionValue: Value 5 for option "simplex_strategy" is above upper bound of 4
Solving

[jajhall]

The primal simplex solver is WIP. Qi wrote it last summer, and I know that there are bugs in it. Please don't invite folk to use it via your wrapper. Once again, we don't have a mechanism for preventing users from setting option values that we need to use for development.

I note the example - which is from issue #280 - and will use it in debugging the primal simplex solver

[jajhall]

Actually, I've just tried to solve the LP given by the MPS file and, whether or not presolve is used, HiGHS returns the correct solution. This is regardless of the simplex options that's chosen - quite simply because no simplex solver needs to be used as the initial point is optimal.

Could you generate one of the random problems and send post it as an MPS file. I don't understand your use of the random function well enough to generate them myself. Thanks

[mckib2]

I will make a note to avoid using this option in production.

This shouldn't be the same example as in #280, I've made a mistake in uploading the incorrect file.

This should be the correct MPS file:
simplex_primal_test.txt

[jajhall]

I'm now reproducing the bug with this MPS file, thanks.

[mckib2]

This is no longer failing on current master, will close