Setting GAM parameters to reduce overfitting

[aboggio-pasqua]

Hi there!

I'm looking for a way to manually define GAM parameters differently than the default (or bigboss) settings, in order to reduce overfitting. I've read that decreasing k (the number of basis functions) and increasing sp (smoothing parameter) could do that, but I don't know how to code this while using the biomod2 package.

Related question: how can I display the GAM parameters used in the bigboss or default strategy?

Many thanks for your help!
Tine

[HeleneBlt]

Hi Tine !

I will first answer your second question :

how can I display the GAM parameters used in the bigboss or default strategy?

You have different ways to do that :

• You can use get_options(myBiomodModelOut) if you already have made some modeling
• You can create a set of options with "bigboss" or "default" like this:

myOpt  <- bm_ModelingOptions(data.type = 'binary',
                             models = c('GAM'),
                             strategy = 'default')

myOpt

• And finally, for bigboss, all the changes parameters are referenced here.

=====================================

Here, an example of how you can modify the parameters of your GAM models:

form.GAM <- bm_MakeFormula(resp.name = myBiomodData@sp.name,
                           expl.var = head(myBiomodData@data.env.var),
                           type = 's_smoother',
                           interaction.level = 0,
                           k = 2)

user.GAM <- list('_for_all_datasets' = list(formula = form.GAM, sp = your_sp_vector))

user.val <- list( GAM.binary.mgcv.gam = user.GAM)

myOpt <- bm_ModelingOptions(data.type = 'binary',
                            models = c('GAM'),
                            strategy = "user.defined",
                            user.val = user.val,
                            user.base = "bigboss",
                            bm.format = myBiomodData)

print(myOpt)

myBiomodModelOut <- BIOMOD_Modeling(bm.format = myBiomodData,
                                    modeling.id = 'Example',
                                    models = c('GAM'),
                                    CV.strategy = 'random',
                                    CV.nb.rep = 3,
                                    CV.perc = 0.7,
                                    OPT.user = myOpt,
                                    metric.eval = c('TSS','ROC'))

You can find a lot of information in the Options vignette or the documentation of bm_MakeFormula.

Hope it helps,
Hélène