[Biomod-commits] Interaction between variables in GBM

[Wilfried Thuiller]

Dear Blaise,

interaction.depth reflects the size of the trees to fit at each  
iteration. One means means only one node and two leaves, and so on.
There is no magic value and one should select the optimal number using  
independant datasets or cross-validation. However, interaction.deph is  
also influencing the learning rate (skrinkage).
Given the BRT is aready quite time-consuming I did not write such a  
function which would find the optimal numbers.
Instead, I followed Elith et al. Journal of Animal Ecology 2009 where  
7 for interaction.depth and skrinkage of 0.001 were the most suitable  
numbers.

If some of you would like to change this parameter, I cold easily add  
it to the Models call so that each of you can change the default value  
for interaction.depth and shrinkage

All the best,
Wilfried




Le 9 nov. 10 à 10:09, Blaise Petitpierre a écrit :

> Dear BIOMODER,
>
> I had a look on the Biomod.Models() function and I was wondering why  
> interaction depth was set to 7 in the GBM function.
>
> gbm(eval(parse(text = paste(SpNames[i],
>                   paste(scopeExpSyst(DataBIOMOD[1:10, 1:NbVar],
>                     "GBM"), collapse = "")))), data =  
> DataBIOMOD[calib.lines,
>                   ], distribution = "bernoulli", var.monotone = rep(0,
>                   length = NbVar), w = Yweights[calib.lines,
>                   i], interaction.depth = 7, shrinkage = 0.001,
>                   bag.fraction = 0.5, train.fraction = 1, verbose = F,
>                   cv.folds = 5)
>
>
>
> Interaction depth is the maximum depth of variable interactions. 1  
> implies an additive model, 2 implies a model with up to 2-way  
> interactions, etc... I was wondering if it was because 7 variables  
> were used in the tutorial but in this case, wouldn't be better to  
> set the interaction depth to the number of variables (NbVar) ?
>
> However I was wondering the effect of interaction depth on  
> overfitting.
>
> Cheers,
> Blaise Petitpierre
> _______________________________________________
> 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
Home page: http://www.will.chez-alice.fr
Website: http://www-leca.ujf-grenoble.fr/equipes/tde.htm

FP6 European MACIS project: http://www.macis-project.net
FP6 European EcoChange project: http://www.ecochange-project.eu






-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/biomod-commits/attachments/20101109/506ec2fe/attachment.htm>