From gael at phys.columbia.edu Fri Jan 22 18:53:40 2010
From: gael at phys.columbia.edu (Gael Reinaudi)
Date: Fri, 22 Jan 2010 12:53:40 -0500
Subject: [Paradiseo-help] continous variable problem
In-Reply-To: <4B4D8D0D.70008@lists.gforge.inria.fr>
References: <4B4CA4C2.1010804@phys.columbia.edu> <4B4CABDF.9060807@lists.gforge.inria.fr>
<4B4CE802.8040302@phys.columbia.edu>
<4B4D8D0D.70008@lists.gforge.inria.fr>
Message-ID: <4B59E624.7080303@phys.columbia.edu>
Hi,
Thank you very much for you answer.
May I ask you if you know anyone who might have implemented a
real-valued hill-climbing optimization? or any real-valued optimization
(I did see the Genetic examples)?
I would love to see an example of that. I feel like I could make
something but I would always question myself if it is the easiest or
most accurate way to do it.
Also, is it an option to take one of the discrete optimization examples
(like the QAD) and to say "anyway I can discretize the values my
real-numbers can take"? eg if I have 3 variables, I would split each
dimension in 1000 values (=1 million possibilities) and then just use
something like the QAD example?
For info, I want to control some parameters of a quantum mechanics
experiment so that it is self optimized.
Thank you very much for having taken the time to answer me
On 1/13/2010 4:06 AM, paradiseo-help wrote:
> Hi,
> I think it's easier to start with the hill-climbing then change the
> representation and operators:
> - the representation of a Solution (maybe you can use a "eoRealVector")
> - move (inherit on "moMove")
> - Incremental evaluation(inherit on "moMoveIncrEval")
> - move Initialization (inherit on "moMoveInit")
> - Next move (inherit on "moMoveNext")
>
> Regards,
> Paradiseo Team.
>
>
>
> Gael Reinaudi a écrit :
>> Hi,
>> thanks for your answer.
>> If I may ask your opinion on one last detail:
>>
>> From wich example should I start and modify to optimize a real
>> valued functor?
>> I mean:
>> - from one of the genetic examples (real valued) then I try to switch
>> the algorithm part,
>> - or, from one of the hill-climbing example (discrete), then I try to
>> define real-value classes and functors ?
>>
>> thanks for your help,
>> Gael
>>
>>
>> On 1/12/2010 12:05 PM, paradiseo-help wrote:
>>> Hi,
>>> Sorry, it not exist a simply way to plug a blackbox. You have to
>>> implement in the incremental evaluation class, a link between the
>>> representation of a solution in ParadisEO and inputs of your
>>> blackbox function (and inversely: output -> ParadisEO representation).
>>>
>>> Best regards,
>>> ParadisEO Team.
>>>
>>> Gael Reinaudi a écrit :
>>>> Hello,
>>>> My name is Gael Reinaudi.
>>>> I am a postdoc at the Columbia University in New York City. I am
>>>> working on a quantum mechanics experiment.
>>>> I have been admiring the ParadisEO framework for days now ! it
>>>> seems to be very complete and nicely object oriented.
>>>>
>>>> I however have a bit of a hard time finding a tutorial that would
>>>> show me :
>>>> How to implement a hill climbing optimization for a simple
>>>> continuous variable problem.
>>>> I have a blackbox function taking N scalar as an input. I am
>>>> would start to change the tutorial about the discrete hill climbing
>>>> (the QAP problem), but I am afraid I would produce a code 10 times
>>>> more complicated than what it should be.
>>>> There must be a simple way (not customized, with less class
>>>> derivation) to plug a real valued functor.
>>>>
>>>> I thank you for your time and hope to here from you soon.
>>>> Gael
>>>>
>>
--
Dr. Gael Reinaudi
tel: (646) 422-9346
Columbia University, Physics Department
538 West 120th Street
Rm 704 Pupin Hall, mail code 126
New York, NY 10027
USA