[Biomod-commits] Weight in ensemble forecasting

[Kumar Mainali]

Hello Wilfried,

Thank you for the explanation. In my case where there are no repetitions,
every model is a final model. Wouldn't you need a way to validate the model
before assigning a weight? That necessitates either cross validation score
or validation score with independent data.

A separate question: I have run BIOMOD in regular R (v 2.14.1) and also in
Revolution R Enterprise. In regular R, it uses just 1/n of the processing
capacity, "n" being the number of processors. It does use different
processors but at different time. Is there a way to use all of the
processing capacity? I encountered the situation even when I had set up the
code for multiple runs. To use all the processing capacity of my machine, I
had to run "n" number of separate R programs.

Many thanks,
Kumar

On Fri, May 11, 2012 at 3:03 AM, Wilfried Thuiller <
wilfried.thuiller at ujf-grenoble.fr> wrote:

> Our mistake, actually not.
>
> The weights are calculated on the final model (100%) if you do not have
> any repetition (like in your case).
>
> We are changing that.
> Priority will be given to "Independent data". Then, cross-validation, then
> final model.
>
> At the moment, we do not check whether or not independent dataset have
> been used.
>
> We'll keep you updated on the package building but everything should work
> back during the day.
>
> Cheers,
> Wilfried
>
>
>
> Le 11 mai 2012 à 08:46, Wilfried Thuiller a écrit :
>
> Dear Kumar,
>
> Ensemble.Forecasting. How the software computes ensemble probabilities is
> not clear in the manual. But somewhere I came across the information that
> the weights are based on cross validated evaluation score. Since I do not
> have such score, the function should not work.
>
>
> No, it works. This is mostly because so few people have independent
> dataset that by default we wrote that the weights are based on the
> cross-validation results. Indeed, when you have independent dataset, the
> weight are calculate based on the scores on the independent evaluation.
> This is something we need to modify in the manual, sorry for the confusion
> it gave you.
>
> Instead, it gives me weight:
>
> $weights
>     ANN CTA    GAM    GBM   GLM MARS FDA     RF SRE
> PA1   0   0 0.1729 0.2766 0.108    0   0 0.4425   0
>
> Where are these weights coming from? These numbers do not match with any
> column in Evaluation.results (after rescaling to make the total =1).
>
> https://r-forge.r-project.org/scm/viewvc.php/pkg/R/Ensemble.Forecasting.R?view=markup&root=biomod&pathrev=168
>
>
>
> It depends on the weighting scheme you use (i.e. decay). Please look at
> the "decay" explanation in the help of Ensemble.Forecasting.
> If you set decay= 'proportional', you should retrieve easily the weights
> with the evaluation scores (as they are proportional).
>
> Hope it helps,
>
> Wilfried
>
>
>
>
> Thanks in advance,
> Kumar
> --
> Section of Integrative Biology
> University of Texas at Austin
> Austin, Texas 78712, USA
>  _______________________________________________
> Biomod-commits mailing list
> Biomod-commits at lists.r-forge.r-project.org
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/biomod-commits
>
>
>   --------------------------
> Dr. Wilfried Thuiller
> Laboratoire d'Ecologie Alpine, UMR CNRS 5553
> Université Joseph Fourier
> BP53, 38041 Grenoble cedex 9, France
> tel: +33 (0)4 76 51 44 97
> fax: +33 (0)4 76 51 42 79
>
> Email: wilfried.thuiller at ujf-grenoble.fr
> Personal website: http://www.will.chez-alice.fr
> Team website: http://www-leca.ujf-grenoble.fr/equipes/emabio.htm
>
> ERC Starting Grant TEEMBIO project:
> http://www.will.chez-alice.fr/Research.html
> FP6 European EcoChange project: http://www.ecochange-project.eu
>
>
>
>
>
>
>
>
>   --------------------------
> Dr. Wilfried Thuiller
> Laboratoire d'Ecologie Alpine, UMR CNRS 5553
> Université Joseph Fourier
> BP53, 38041 Grenoble cedex 9, France
> tel: +33 (0)4 76 51 44 97
> fax: +33 (0)4 76 51 42 79
>
> Email: wilfried.thuiller at ujf-grenoble.fr
> Personal website: http://www.will.chez-alice.fr
> Team website: http://www-leca.ujf-grenoble.fr/equipes/emabio.htm
>
> ERC Starting Grant TEEMBIO project:
> http://www.will.chez-alice.fr/Research.html
> FP6 European EcoChange project: http://www.ecochange-project.eu
>
>
>
>
>
>
>
>


-- 
Section of Integrative Biology
University of Texas at Austin
Austin, Texas 78712, USA
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/biomod-commits/attachments/20120511/578003d1/attachment.html>