[R] errorest: Error in cv.numeric()

[Torsten Hothorn]

>
> Hi,
>
> I am trying to get an error estimation
> for a classification done using lda.
>
> The examples work fine, however I don't get
> my own code to work.
>
> The data is in object d
>
> 	> d
> 	    class hydrophobicity    charge     geometry
> 	1       2      6490.0400 1434.9700    610.99902
> 	2       2      1602.0601  400.6030  -5824.00000
> 	3       2       969.0060  260.1360   -415.00000
> 	4       1       527.2310  158.7020    -22.00010
> 	...
> 	180     1       299.5190   85.9201   -680.00000
> 	181     2      1385.6801  298.8360   -353.00000
> 	182     1       428.8740  130.8020   -328.00000
> 	183     1       287.5540   98.0767     34.00000
>

the class of `d' is "numeric" and should be "factor", which is implicitly
assumed for classification problems (lda kindly operates on it but, for
example rpart, won't).

Best,

Torsten

> Since predict.lda does not return simply the classification
> it is wrapped, as in the docs:
>
> 	mypredict.lda <- function(object, newdata) predict(object, newdata = newdata)$class
>
> In trying errorest() I get the message
>
> 	> errorest(class ~ hydrophobicity + charge, data=d, model=lda, predict=mypredict.lda)
> 	Error in cv.numeric(y, formula, data, model = model, predict = predict,  :
> 	        predict does not return numerical values
>
> even though a "manual" lda seems to provide the correct types:
>
> 	> l <- lda(class ~ hydrophobicity + charge, data=d)
> 	> mypredict.lda(l,d)
> 	  [1] 2 2 1 1 2 2 1 2 1 2 2 1 2 1 1 2 1 1 2 1 1 2 2 1 2 1 1 2 1 1 1 1 1 2 1 1 1
> 	...
> 	[149] 1 2 1 1 1 1 1 1 2 1 2 1 2 1 1 1 1 2 1 2 2 2 1 2 1 1 1 1 1 1 1 1 1 1 1
> 	Levels: 1 2
>
> 	> typeof(mypredict.lda(l,d))
> 	[1] "integer"
>
> So what did I miss, I am fairly new to programming in R/S,
> so I might have missed some conventions and/or conversions.
>
> Software used is
>
> 	R Version 1.6.2  (2003-01-10) / Linux
> 	ipred_0.6-14.tar.gz
>
> Yours,
> Steffen
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help
>
>