[Biomod-commits] Error in newx$cptable[max(keep), 1L] <- cp : subscript out of bounds

[jason b mackenzie]

Hi Wilfried

Thanks for the suggestions to drop low performing techniques (CTA, ANN and SRE) and redundant evaluation methods (KAPPA), and to choose another pseudo absence strategy (random). I look forward to reading the paper, and that's all great practical guidance. 

I ran all the models mostly for exploratory purposes, with the assumption that i could identify the worst performers with evaluation scores and drop them from Ensemble.Forecasting ( ) with flags such as weight.method='proportional', decay = 1.6, qual.th = 0.5. With that said, it makes since to drop weak approaches up front, so thanks again for the advice.

In case there are other methods producing curious errors besides CTA, a different species fell over yesterday on FDA, with the following error:
Model=Flexible Discriminant Analysis 
Error in family$linkfun(mustart) : 
  Argument mu must be a nonempty numeric vector

Cheers, 
Jason



On Feb 10, 2012, at 6:11 PM, Wilfried Thuiller wrote:

> Dear Jason,
> 
> The problem is related to the classification tree. It seems the tree has no branches and BIOMOD does not manage to get the optimal number of leaves. 
> My take on that is that your species is very difficult to model for CTA and it does not manage to find any good combination of variables explaining the distribution. 
> Perhaps we need to make an error trap around this call as this is not the first time we got this error, which is not a BIOMOD bug, but rather the failure of CTA to find an optimal number of leaves. 
> I will suggest removing the model for this species.
> 
> Looking at the code you sent with this message and the other one, I would also suggest:
> 
> Removing some techniques. Some generally performed quite poorly compared to the other ones. CTA and ANN are the weakest ones. SRE is more like a null model, the most liberal. I would drop the three. 
> Same for evaluation, you do not need to use both TSS and KAPPA, they are quite redundant. Just keep TSS, less influenced by prevalence and ROC. 
> 
> I would recommend using "random" for pseudo absence instead of SRE or circle which tends to over fit a bit the data.
> I take the liberty to send you this paper in press with Methods in Ecology and Evolution that discuss the different strategies for selecting pseudo-absence:
> PDF
> 
> Hope it helps,
> 
> Wilfried
> 
> 
> 
> 
> Le 7 févr. 2012 à 05:06, jason b mackenzie a écrit :
> 
>> Dear all,
>> 
>> I recently had a BIOMOD run crash mid-stream after 20+ species appear to have run successfully. The error message is pasted below. Any ideas or workarounds? 
>> 
>> ...
>> #####			 ERFA2 			#####
>> #####		   pseudo-absence run 1        		#####
>> Model=Artificial Neural Network 
>> 	 3 Fold Cross Validation + 3 Repetitions 
>> Calibration and evaluation phase: Nb of cross-validations:  3 
>> Evaluating Predictor Contributions in  ANN ... 
>> Model=Classification tree 
>> 	 50 Fold Cross-Validation 
>> Error in newx$cptable[max(keep), 1L] <- cp : subscript out of bounds
>> > warnings()
>> Warning messages:
>> 1: In predict.lm(object, newdata, se.fit, scale = 1, type = ifelse(type ==  ... :
>>   prediction from a rank-deficient fit may be misleading
>> 2: In cor(g.pred[, ], as.integer(.Rescaler4(as.numeric(predict(model.sp,  ... :
>>   the standard deviation is zero
>> ...
>> 
>> 
>> FYI, not sure if this matters, but my starting conditions were set as follows:
>> 
>> Initial.State(Response=SP.ENV[,12:75], Explanatory=SP.ENV[,4:11], IndependentResponse=NULL, 
>> 		IndependentExplanatory=NULL)
>> 
>> Models(GLM = FALSE, GAM = TRUE, Spline=4, GBM = TRUE, No.trees = 3000, CTA = TRUE, CV.tree = 50, ANN = TRUE, CV.ann = 3, 
>> 		SRE = TRUE, quant = 0.025, FDA = TRUE, MARS = TRUE, RF = TRUE, NbRunEval = 3, DataSplit = 70, 
>> 		Yweights = NULL, Roc = TRUE, Optimized.Threshold.Roc = TRUE, Kappa = TRUE, TSS = TRUE, 
>> 		KeepPredIndependent = FALSE, VarImport = 5, NbRepPA = 2, strategy = "sre", nb.absences = 1000)
>> 
>> 
>> Thanks in advance for any help!
>> 
>> Cheers, 
>> Jason
>> 
>> 
>> --
>> 
>> jason b mackenzie
>> 80 banambila street
>> aranda, act 2614
>> 
>> 0447 002 629  [mobile]
>> jasonbmackenzie    [skype]
>> 
>> 
>> 
>> 
>> _______________________________________________
>> 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
> 
> 
> 
> 
> 
> 
> 


--

jason b mackenzie
80 banambila street
aranda, act 2614

0447 002 629  [mobile]
jasonbmackenzie    [skype]




-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/biomod-commits/attachments/20120210/643a6c44/attachment-0001.html>