[R] Cv.glment question -- why giving me an error

[Patrick Breheny]

On 08/11/2011 01:01 PM, Andra Isan wrote:
> Hi All,
> I am trying to run cv.glmnet(x,y,family="multinomial", nfolds =4) and I only have 8 observations and the number of features I have is 1000, so my x matrix is 8 by 1000 and when I run the following, I get this error, I am not sure what is causing this problem.
> Error in predmat[which, , seq(nlami)] = preds :   number of items to replace is not a multiple of replacement length
> Can it be because the number of observations is small?
> Thanks a lot,Andra
> 	[[alternative HTML version deleted]]
>

1) My guess is that you do not have an observation in each class in each 
fold.  It would be rather difficult to estimate the probability of 
belonging to a class without any data.

2) Do you really hope to obtain a meaningful fit for a multinomial model 
with only 8 observations?  How many classes do you have?

-- 
Patrick Breheny
Assistant Professor
Department of Biostatistics
Department of Statistics
University of Kentucky