commons-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Glautier S.P." <S.P.Glaut...@soton.ac.uk>
Subject RE: [math]
Date Wed, 27 Feb 2013 14:05:32 GMT
Thanks Gilles, I'm using 3.0, will try again with latest.

I'm minimising and if updating doesn't fix it I'll send some more information.

Steve.

>-----Original Message-----
>From: Gilles [mailto:gilles@harfang.homelinux.org] 
>Sent: 27 February 2013 13:06
>To: user@commons.apache.org
>Subject: Re: [math]
>
>Hello.
>
>On Wed, 27 Feb 2013 12:23:44 +0000, Glautier S.P. wrote:
>> Dear list, I'd be grateful for any help on the following:
>>
>> I am using the CMAESOptimizer in the java commons.math3 library to 
>> generate a best fit for a model of learning to some experimental data.
>> When I take the best parameters generated by the optimisation and plug 
>> them into the model, the fit between the model and the data doesn't 
>> match that which I expect on the basis of the optimisation output. In 
>> the cases I have checked the optimiser output best fit (the value part 
>> of the PointValuePair that comes at the end of the optimisation) is 
>> slightly larger than the value I get when I run the model using the 
>> parameters from the optimiser output (the point part of the 
>> PointValuePair that comes at the end of the optimisation).
>>
>> I have spent quite a long time checking my code and I can't find what 
>> the problem is but it occurred to me that there might be something 
>> about the optimisation output that I haven't understood properly. Am I 
>> correct to assuming that the parameters produced at the end of the 
>> optimisation are those that produced the associated best fit value?
>
>Yes, that should be so.
>
>Which version of Commons Math are you using?
>
>[To figure out what could have happened, it would be easier to have the actual numbers
for the parameters and associated cost and the function to be optimized (minimized or maximized?).]
>
>
>Best regards,
>Gilles
Mime
View raw message