You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@mahout.apache.org by Lance Norskog <go...@gmail.com> on 2011/12/30 03:32:12 UTC
AdaptiveLogisticRegression for multinomial values?
AdaptiveLogisticRegression includes the following note about
beingworthwhile for binary values only. Is this still true, or are
therealternative crossfold learners for multinomial data?
org.apache.mahout.classifier.sgd.AdaptiveLogisticRegression----------
* The fitness used here is AUC. Another alternative would be to try
log-likelihood, but it is much * easier to get bogus values of
log-likelihood than with AUC and the results seem to accord pretty *
well. It would be nice to allow the fitness function to be pluggable.
This use of AUC means that * AdaptiveLogisticRegression is mostly
suited for binary target variables. This will be fixed * before long
by extending OnlineAuc to handle non-binary cases or by using a
different fitness * value in non-binary cases.---------------
--
Lance Norskog
goksron@gmail.com
Re: AdaptiveLogisticRegression for multinomial values?
Posted by Ted Dunning <te...@gmail.com>.
This is still true.
The limitation is not with the cross-fold learner, but with the adaptive
learners fitness function.
You should be able to get useful work done with the non-adaptive version,
OnlineLogisticRegression, which doesn't require AUC to work.
On Thu, Dec 29, 2011 at 6:32 PM, Lance Norskog <go...@gmail.com> wrote:
> AdaptiveLogisticRegression includes the following note about
> beingworthwhile for binary values only. Is this still true, or are
> there alternative crossfold learners for multinomial data?
>
> org.apache.mahout.classifier.sgd.AdaptiveLogisticRegression
----------
> * The fitness used here is AUC. Another alternative would be to
> try log-likelihood, but it is much
* easier to get bogus values of log-likelihood than with AUC and the
> results seem to accord pretty
* well. It would be nice to allow the fitness function to be pluggable.
> This use of AUC means that
* AdaptiveLogisticRegression is mostly suited for binary target variables.
> This will be fixed
* before long by extending OnlineAuc to handle non-binary cases or by using
> a different fitness
* value in non-binary cases.
---------------
>
> --
> Lance Norskog
> goksron@gmail.com
>