Tuning CPLEX

[pluis1908]

Hi guys!!

I'm using the automatic tuning tool of CPLEX 12.4 to determinate the most adequate setting for a MIP model. 

However, I need to solve several instances (possibly of different sizes) and I do not want to get the best parameter configuration for every single instance, but for the whole set of instances. 

Does anyone know how I can run these several instances and get the best configuration for the whole set? 

I'll be grateful for your help. 

Best,

Pedro. 

[Robert Fourer]

AMPL can only work with CPLEX to tune on one instance at a time. However if you tune to several instances, you can look for settings that CPLEX recommends for all or most instances, and use those for your runs.

In my experience the results from the tuning tool should be regarded only as suggestions, which should be tested by further experiments. Often it turns out that some of the suggestions make little difference, while others are useful across many instances.

Bob Fourer
am...@googlegroups.com

=======

[pluis1908]

Dear Bob,

Thanks so much for your comments. 

It looks a good idea to tune several instances, one at a time, and then look for the most common settings for the whole set. I'll do my runs in that way. 

Best regards,

Pedro. 

[juanman...@gmail.com]

Hi. I have a big MIP model.
In AMPL how can I use the automatic tuning tool. Can you give an example please. Thanks...

[Robert Fourer]

For CPLEX (mentioned in your subject line), you can specify, say,

option cplex_options 'tunefile=tune.txt';

Then you will get some output showing the progress of the tuning run, and when the run is finished you will find the recommended cplex_options settings in the file tune.txt. You can also find at http://ampl.com/products/solvers/solvers-we-sell/cplex/options/ the instructions for other cplex_options settings that influence the tuning run; look for entries that begin with "pretune" or "tune".

Other MIP solvers have tuning features that work similarly, but with somewhat different details and setting names as described in their option listings.

Bob Fourer
am...@googlegroups.com

=======

From: am...@googlegroups.com [mailto:am...@googlegroups.com] On Behalf Of juanman...@gmail.com
Sent: Tuesday, May 30, 2017 9:41 PM
To: AMPL Modeling Language

[juan manuel home ortiz]

Dear Bob,

Thanks so much for your comments.

--
You received this message because you are subscribed to a topic in the Google Groups "AMPL Modeling Language" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/ampl/XmlqQOXS1gE/unsubscribe.
To unsubscribe from this group and all its topics, send an email to ampl+unsubscribe@googlegroups.com.
To post to this group, send email to am...@googlegroups.com.
Visit this group at https://groups.google.com/group/ampl.
For more options, visit https://groups.google.com/d/optout.

--

Juan Manuel Home Ortiz

Ph.D Student.

Universidade Estadual Paulista.

UNESP - Ilha Solteira - SP