[R] changing the loss function in the logistic regression?

[Frank E Harrell Jr]

Michael wrote:
> Hi Frank,
> 
> Thanks for your help!
> 
> I want to incorporate lift score as the optimization objective. How to
> do that in logistic regression?
> 
> Thanks!

Please re-read my note.

Models should be fitted using proper scoring rules.  Otherwise the 
resulting fit is bogus.

Thanks
Frank

> 
> On Fri, Jun 26, 2009 at 7:47 AM, Frank E Harrell
> Jr<f.harrell at vanderbilt.edu> wrote:
>> Michael wrote:
>>> Hi all,
>>>
>>> Is there a way to change the loss function in the logistic regression?
>>> Or we could provide a customized loss function in the logistic
>>> regression so we could use that loss function in the Cross Validation
>>> in logistic regression?
>>>
>>> Thanks a lot!
>> The goal is to use a loss function that yields optimality, with a sensible
>> definition of optimality.  For many purposes, maximum likelihood or
>> penalized maximum likelihood is optimum.  So don't change the optimality
>> criteria just because you are cross-validating a different measure.
>>
>> By the way, it's often not a good idea to cross-validate a different
>> measure.  At least the accuracy index should be information-preserving.
>>  Deviance, log-likelihood, and AIC are your friends.
>>
>> Frank
>>
>> --
>> Frank E Harrell Jr   Professor and Chair           School of Medicine
>>                     Department of Biostatistics   Vanderbilt University
>>
> 


-- 
Frank E Harrell Jr   Professor and Chair           School of Medicine
                      Department of Biostatistics   Vanderbilt University