[R] SVM Prediction and Plot

[Steve Lianoglou]

Hi,

On Wed, Feb 2, 2011 at 1:40 PM, Brian <cttobin at hotmail.com> wrote:
>
> There are a good few blanks in some variables in the data. There were in the
> training set too. Is that a problem?

Well, it looks like it's the root cause of what prompted you to post
to R-help ... whether or not it's *really* a problem is another issue,
I guess ...

> I don't have any na.action in my svm
> call either.

Makes sense ... the SVM is just skipping these examples and not using
them for training. The default setting for na.action is na.omit, which
just removes these cases from training.

> So na.omit=na.fail causes this error:
> Error in na.fail.default(newdata) : missing values in object

Bingo.

> I tried the matrix.
>>SvmPred = predict(Svm, as.matrix(BankTest[,-1]))
> Error in scale.default(newdata[, object$scaled, drop = FALSE], center =
> object$x.scale$"scaled:center",  :
>  length of 'center' must equal the number of columns of 'x'
>
> Also is.matrix fails for BankTest so it isn't actually a matrix. Not sure it
> matters though. The prediction is actually working somewhat, but it only
> predicts for 2282 cases instead of 2598.

We know why now. Please read the help for the predict.svm function:

R> ?predict.svm

Specifically the documentation for the `na.action` Argument. Perhaps
it would be more friendly if predict just returned an NA in these
situations, but I guess this is how it works for the time being.

Anyway, it looks like this explains why you have a different number of
predictions than you have observations in your BankTest.

-steve

-- 
Steve Lianoglou
Graduate Student: Computational Systems Biology
 | Memorial Sloan-Kettering Cancer Center
 | Weill Medical College of Cornell University
Contact Info: http://cbio.mskcc.org/~lianos/contact