[Biomod-commits] Warning in BIOMOD_FormatingData

[Damien Georges]

Dear Jesus,

I agree with your comment, that's why we don't want to give a mean or a 
median of variables importance of all single models selected as variable 
importance...

As for ensemble models evaluation, we have to consider ensemble models 
as "true" models.. So we have to apply the same methodology than for the 
single models..
We must randomise a variable, calculate all single models predictions 
with this 'new set of data', combine predictions to build our ensemble 
model, calculate the correlation between the 'true' predictions of our 
'ensemble model' and the ones with a randomised variable....Do it 
several times for all variables and all models.... It will be really 
time consuming but might be interesting in some cases.

I guess that calculating variables importance according to methodology 
describes above will give you a quite interpretable value...

Cheers,

Damien.


On 13/09/2012 09:57, Jesus wrote:
>
>
> On 13 September 2012 09:49, Damien Georges <damien.georges2 at gmail.com 
> <mailto:damien.georges2 at gmail.com>> wrote:
>
>     Dear Pablo,
>
>     There is no way to extract variables importance from a Ensemble
>     model because it's not calculated yet.. We are thinking about the
>     way to implement it  because it's a really time consuming step..
>     Concerning the response plots, the problem is the same..
>     I will tell you when this functionalities will be ready.
>
>     Cheers,
>
>     Damien.
>     Dear Damien, thanks for the responses to this threat. I however
>     have a question/observation about the variables importance in a
>     ensemble of models.  How do you really interpret the variable
>     importance coming from an ensemble of models? Imaging two
>     algorithms gave you high AUC values so they are considered good
>     models, but choose different environmental variables, what would
>     be the variable importance result of the ensemble? I find the
>     interpretation of this a bit tricky. and not that straight forward.
>
>
> Jesus
>
>     On 12/09/2012 17:59, Pablo Riul wrote:
>>     Dear Damien,
>>
>>     Ok, thanks! Just another thing... I'm not sure if you received my
>>     anterior message.  I tried to get the var importances of a
>>     consensus model using something like getModelsVarImport() to an
>>     object created by the BIOMOD_EnsembleModeling() function but it
>>     was not possible. Is there any way to get these information?
>>     Also, will be possible to get the response plots for this same
>>     object?
>>
>>     Cheers,
>>     Pablo Riul
>>
>>     Departamento de Engenharia e Meio Ambiente
>>     Centro de Ciências Aplicadas e Educação
>>     Universidade Federal da Paraíba - Campus IV
>>     R: Mangueira s/n, Centro CEP: 58.297-000
>>     Rio Tinto - Paraíba - Brasil
>>
>>     ------------------------------------------------------------------------
>>     *De:* Damien Georges <damien.georges2 at gmail.com>
>>     <mailto:damien.georges2 at gmail.com>
>>     *Para:* biomod-commits at lists.r-forge.r-project.org
>>     <mailto:biomod-commits at lists.r-forge.r-project.org>; Pablo Riul
>>     <pabloriul at yahoo.com.br> <mailto:pabloriul at yahoo.com.br>
>>     *Enviadas:* Quarta-feira, 12 de Setembro de 2012 10:52
>>     *Assunto:* Re: [Biomod-commits] Warning in BIOMOD_FormatingData
>>
>>     Dear Pablo,
>>
>>     Don't worry about it.. It's just a warning indicating that some
>>     function of raster package have changed of names..
>>     I took this into account in the last biomod2 version so warnings
>>     will disappear when you'll update the package .
>>
>>     Best,
>>
>>     Damien
>>
>>     On 12/09/2012 14:46, Pablo Riul wrote:
>>>     Dear list,
>>>
>>>     I updated my packages and I'm now using raster 2.0-12. When I
>>>     created myBiomodData using the BIOMOD_FormatingData function I
>>>     received the warning below:
>>>
>>>     >myBiomodData <- BIOMOD_FormatingData(resp.var = myResp,expl.var
>>>     = myExpl,resp.xy = myRespXY,resp.name <http://resp.name> =
>>>     myRespName,PA.nb.rep = 1,PA.nb.absences = 1000,PA.strategy =
>>>     'random')
>>>
>>>     -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= Mhisp
>>>     Data Formating
>>>     -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
>>>
>>>           ! No data has been set aside for modeling evaluation
>>>        > Pseudo Absences Selection checkings...
>>>        > random pseudo absences selection
>>>        > Pseudo absences are selected in explanatory variables
>>>     -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
>>>     Done
>>>     -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
>>>     Warning message:
>>>     In .local(x, rcl, ...) :
>>>       raster function "reclass" has been deprecated and will be
>>>     removed from the package;
>>>     Please use "reclassify" instead
>>>
>>>
>>>     Its not a problem now, but maybe in next updates of package
>>>     raster the reclass function may be not available anymore.
>>>
>>>     Cheers,
>>>     Pablo Riul
>>>
>>>     Departamento de Engenharia e Meio Ambiente
>>>     Centro de Ciências Aplicadas e Educação
>>>     Universidade Federal da Paraíba - Campus IV
>>>     R: Mangueira s/n, Centro CEP: 58.297-000
>>>     Rio Tinto - Paraíba - Brasil
>>>
>>>
>>>     _______________________________________________
>>>     Biomod-commits mailing list
>>>     Biomod-commits at lists.r-forge.r-project.org  <mailto:Biomod-commits at lists.r-forge.r-project.org>
>>>     https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/biomod-commits
>>
>>
>>
>
>
>     _______________________________________________
>     Biomod-commits mailing list
>     Biomod-commits at lists.r-forge.r-project.org
>     <mailto:Biomod-commits at lists.r-forge.r-project.org>
>     https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/biomod-commits
>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/biomod-commits/attachments/20120913/ffe45d75/attachment.html>