R help - Jun 2011 - predict with model (rms package)

Dear R-help,

In the rms package, I have fitted an ols model with a variable
represented as a restricted cubic spline, with the knot locations
specified as a previously defined vector. When I save the model object
and open it in another workspace which does not contain the vector of
knot locations, I get an error message if I try to predict with that
model. This also happens if only one workspace is used, but the vector
of knot locations is removed:

library(rms)
set.seed(1)
x <- rnorm(100)
y <- 1 + x + x^2 + rnorm(100)

x.knots <- quantile(x, c(0.2, 0.5, 0.8))
ols1 <- ols(y ~ rcs(x, x.knots))

predict(ols1, data.frame(x = 0))  # This works
rm(x.knots)
predict(ols1, data.frame(x = 0))  # Gives error

The first predict gives
        1
0.8340293

while the second predict gives
Error in rcs(x, x.knots) : object 'x.knots' not found

The same error happens if x.knots is simply defined as a vector like
c(-1, 0, 1) (i.e. not using quantile). Is this the intended behaviour?
The requirement that x.knots be in the workspace seems strange, given
that the knot locations are stored in ols1$Design$parms.

Thanks for any help you can give.

Mark Seeto
National Acoustic Laboratories, Australia

This is a consequence of predict.ols calling predictrms which relies on
model.frame which re-issues the expression of x.knots.  You would have the
same problem if using update(fit object) in another session.  For that
reason you have to keep an external knots vector available in your
environment.

Frank


Mark Seeto wrote:
> 
> Dear R-help,
> 
> In the rms package, I have fitted an ols model with a variable
> represented as a restricted cubic spline, with the knot locations
> specified as a previously defined vector. When I save the model object
> and open it in another workspace which does not contain the vector of
> knot locations, I get an error message if I try to predict with that
> model. This also happens if only one workspace is used, but the vector
> of knot locations is removed:
> 
> library(rms)
> set.seed(1)
> x <- rnorm(100)
> y <- 1 + x + x^2 + rnorm(100)
> 
> x.knots <- quantile(x, c(0.2, 0.5, 0.8))
> ols1 <- ols(y ~ rcs(x, x.knots))
> 
> predict(ols1, data.frame(x = 0))  # This works
> rm(x.knots)
> predict(ols1, data.frame(x = 0))  # Gives error
> 
> The first predict gives
>         1
> 0.8340293
> 
> while the second predict gives
> Error in rcs(x, x.knots) : object 'x.knots' not found
> 
> The same error happens if x.knots is simply defined as a vector like
> c(-1, 0, 1) (i.e. not using quantile). Is this the intended behaviour?
> The requirement that x.knots be in the workspace seems strange, given
> that the knot locations are stored in ols1$Design$parms.
> 
> Thanks for any help you can give.
> 
> Mark Seeto
> National Acoustic Laboratories, Australia
> 
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
> 

-----
Frank Harrell
Department of Biostatistics, Vanderbilt University
--
View this message in context:
http://r.789695.n4.nabble.com/predict-with-model-rms-package-tp3581229p3582758.html
Sent from the R help mailing list archive at Nabble.com.