[R-sig-ME] MCMCglmm predictions using new data error

[Alberto Gallano]

Hi Sven,

thanks so much. That works. Do you know if there is any documentation as to
how to set up new data for the MCMCglmm predict method? Adding a constant
response vector to the prediction grid is different to every other R
predict method (that I know of).

One further problem i'm having is that I actually have random effects, and
now the predict function throws an error when these are not entered into
the prediction grid. It's unclear to me how they should be entered - using
a constant doesn't work. The number of random effect grouping levels may
not equal the dimensions of the rest of the prediction grid, so just adding
a vector with each unique grouping level won't work here either. Any
thoughts would be much appreciated. See "pred_grid2" at bottom of code
below:


#
----------------------------------------------------------------------------------
library(MCMCglmm)

set.seed(123)

dat <- data.frame(
    y = rnorm(100),
    x = rnorm(100),
    id = factor(rep(paste("ID", 1:20), 5))
    )

fit <- MCMCglmm(
    fixed = y ~ x,
    random = ~ id,
    rcov = ~ units,
    data = dat,
    family = "gaussian",
    pr = TRUE, pl = TRUE,
    saveX = TRUE,  saveZ = TRUE,
    nitt = 1.3e+4, thin = 10, burnin = 3e+3
)


pred_grid1 <- with(dat,
    data.frame(
    y = 0,
    x = seq(-1, 1, length.out = 30)
    ))

predict(fit, newdata = pred_grid1)

# > Error in buildZ(rmodel.terms[r], data = data, nginverse =
names(ginverse)) : object id not found


pred_grid2 <- with(dat,
    data.frame(
    y = 0,
    x = seq(-1, 1, length.out = 30),
    id = 0
    ))

predict(fit, newdata = pred_grid2)

# > Error in MCMCglmm(fixed = object$Fixed$formula, random =
object$Random$formula,  : trying to get slot "Dim" from an object of a
basic class ("numeric") with no slots
#
----------------------------------------------------------------------------------

On Thu, Jan 7, 2016 at 3:06 PM, Sven E. Templer <sven.templer at gmail.com>
wrote:

> Hi Alberto,
>
> avoid the error from predict by running
>
> pred_grid$y <- 0
>
> before
>
> predict(fit, newdata = pred_grid)
>
> Best,
> Sven
>
> > On 06 Jan 2016, at 23:39, Alberto Gallano <alberto.gc8 at gmail.com> wrote:
> >
> > I'm trying to make predictions from an MCMCglmm model using new data.
> This
> > is a feature that was recently added to the predict.MCMCglmm function
> > (version 2.22). However, when I set things up as I would for other
> predict
> > methods, I get the following error:
> >
> >> Error in eval(expr, envir, enclos) : object 'y' not found
> >
> > where 'y' is my response vector. I'm including a simplified replicable
> > example below. Is the set up of the prediction grid different for the
> > MCMCglmm predict method compared with other methods?
> >
> > best,
> > Alberto
> >
> > #
> >
> ----------------------------------------------------------------------------------
> > library(MCMCglmm)
> >
> > set.seed(123)
> >
> > dat <- data.frame(x = rnorm(100), y = rnorm(100))
> >
> >
> > fit <- MCMCglmm(
> >
> >    fixed = y ~ x,
> >
> >    rcov = ~ units,
> >
> >    data = dat,
> >
> >    family = "gaussian",
> >
> >    pr = TRUE, pl = TRUE,
> >
> >    saveX = TRUE,  saveZ = TRUE,
> >
> >    nitt = 1.3e+4, thin = 10, burnin = 3e+3
> >
> > )
> >
> >
> > pred_grid <- data.frame(x = seq(-1, 1, length.out = 30))
> >
> >
> > predict(fit, newdata = pred_grid)
> >
> > #
> >
> ----------------------------------------------------------------------------------
> >
> >       [[alternative HTML version deleted]]
> >
> > _______________________________________________
> > R-sig-mixed-models at r-project.org mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
>

	[[alternative HTML version deleted]]