[R] Error message in function mars() in package mda

[Martin Maechler]

>>>>> "Jude" == Jude Ryan <juderyan at tcindex.com>
>>>>>     on Tue, 10 Aug 2004 15:56:38 -0400 writes:

    Jude> Hi, I am using function mars() in package mda to find
    Jude> knots in a whole bunch of predictor variables. I hope
    Jude> to be able to replicate all or some of the basis
    Jude> functions that the MARS software from Salford Systems
    Jude> creates.  When I ran mars() on a small dataset, I was
    Jude> able to get the knots.  However, when I tried running
    Jude> mars() on a larger dataset (145 predictor variables),
    Jude> for a different project, I get the following error
    Jude> message:

    >> fit1 <- mars(disney2[,-146], disney2[,146])
    Jude> Error in mars(disney2[, -146], disney2[, 146]) :
    Jude> NA/NaN/Inf in foreign function call (arg 5)
    Jude> In addition: Warning messages:
    Jude> 1: NAs introduced by coercion
    Jude> 2: NAs introduced by coercion
    >> 

    Jude> Does arg 5 refer to the 5th column in my dataset? 
no.

    Jude> This seems to be a data problem, is this correct?

we cannot know with the little information you give.
Please read (and follow!!) the posting guide!
-> reproducible example !
-> contact the maintainer of the package in question!

The following reproducible example doesn't show the problem you
mentioned:

> library(mda)
> set.seed(101)
> X <- matrix(rnorm(200 * 150), 200, 150)
> fX <- mars(X[,-146], X[,146])
> str(fX)
List of 15
 $ call          : language mars(x = X[, -146], y = X[, 146])
 $ all.terms     : int [1:93] 1 2 3 4 5 6 7 8 9 10 ...
 $ selected.terms: int [1:25] 1 2 3 4 5 10 12 13 14 16 ...
 $ penalty       : num 2
 $ degree        : num 1
 $ nk            : num 299
 $ thresh        : num 0.001
 $ gcv           : num 0.72
 $ factor        : num [1:99, 1:149] 0 0 0 0 0 0 0 0 0 0 ...
  ..- attr(*, "dimnames")=List of 2
  .. ..$ : NULL
  .. ..$ : NULL
 $ cuts          : num [1:99, 1:149] 0 0 0 0 0 0 0 0 0 0 ...
 $ residuals     : num [1:200, 1]  0.789  0.111 -0.212  1.017 -0.391 ...
 $ fitted.values : num [1:200, 1] -0.216 -0.249 -0.116 -0.274 -0.353 ...
 $ lenb          : int 99
 $ coefficients  : num [1:25, 1]  0.696  0.491  1.013 -1.043 -0.409 ...
 $ x             : num [1:200, 1:25] 1 1 1 1 1 1 1 1 1 1 ...
 - attr(*, "class")= chr "mars"
> 

BTW (as a note to the maintainer of "mda"):

  print.mars()  {and maybe summary.mars()} is severly lacking...

    Jude> Are there any other functions in R that will give me
    Jude> the knots for a set of predictor variables?

maybe, quite probably. This question is too vague ["the" knots?]


Regards,
Martin Maechler