Originally posted by: Hlads

I am using Python API for IBM CPLEX. I am quite new in CPLEX.

In my program I iteratively set to zero bounds of different combinations of variables and solve the problem for each such combination. In some cases I get unbounded solution (while before setting to zero this bounds problem was bounded).

(To avoid infeasibility errors I turn off scaling

problem.parameters.read.scale.set(-1))

Example of output data (combination of variables/objective/status):

When I check this combinations "by hand" I get bounded solution with different value of objective.

What is the reason of such behaviour? How can I solve this problem?

Is it connected somehow with previous solving steps (before setting to zero bounds of variables)?

This may or may not be helpful.  Did you see Coping with an ill-conditioned problem or handling unscaled infeasibilities in the documentation?

Also, are you able to export the model before calling solve() (preferably in .sav.gz format) and attach it here (for a case where you think you're getting the wrong results)?  This may allow others to help you more quickly.  Do you get the same results when running the exported model in the interactive?

Originally posted by: Hlads

Previously I had this problem with scaling, in this case I got status "optimal_infeasible", that's why I turn off scaling.

But now it is seems that it is a different problem.

When I do the same interactively reading problem from .sav file I get different results. Status - optimal.

Unfortunately I can't upload my problem, as this is unpublished data.

I have found the following in documentation:

Re-solving

You may re-solve the problem by reissuing the optimize command. CPLEX restarts

the solution process from the previous optimal basis, and thus requires zero

iterations. If you do not wish to restart the problem from an advanced basis, use

the set advance command to turn off the advanced start indicator.

Remember that a problem must be present in memory (entered via the enter

command or read from a file) before you issue the optimize command.

How can I control this parameters in Python API?

By default, this parameter is already "on".

You can set the advanced start switch parameter in the Python API, like so:

>>> import cplex
>>> cpx = cplex.Cplex()
>>> cpx.parameters.advance.values[cpx.parameters.advance.get()]
'standard'
>>> cpx.parameters.advance.set(cpx.parameters.advance.values.none)
>>> # run the following to see the help:
... 
>>> help(cpx.parameters.advance.values)

Originally posted by: Hlads

Thank you!