Skip to content

Data Handling within nlme library

2 messages · biii m@iii@g oii de@@ey@ws, Phillip Alday

Original
biii m@iii@g oii de@@ey@ws
# 
Hello,

 

I'm having an issue with data handling within the nlme library.
Specifically, when I fit a gnls model within one function, I cannot update
that model within another function with new data.  Are there any ways to
give data within a function and have the model work outside?  An example of
what I'm trying to do is below in the first example:

 

``` r

library(nlme)

 

make_model <- function() {

  my_soybean <- Soybean

  gnls(weight ~ SSlogis(Time, Asym, xmid, scal), data=my_soybean, weights =
varPower())

}

 

my_model <- make_model()

#> Error in stats::nls(formula = weight ~ SSlogis(Time, Asym, xmid, scal), :
object 'my_soybean' not found

# Issue: my_soybean is in the function environment but not found during
estimation

 

make_model <- function() {

  time <- Soybean$Time

  weight <- Soybean$weight

  gnls(

    weight ~ Asym/(1+exp((xmid-time)/scal)),

    start=c(Asym=30, xmid=40, scal=10),

    weights = varPower()

  )

}

 

my_model <- make_model()

#> Error in terms.formula(formula, data = data): '.' in formula and no
'data' argument

# Issue: No idea where the error is coming from (making a guess that it is
the weights argument)

 

make_model <- function() {

  time <- Soybean$Time

  weight <- Soybean$weight

  gnls(

    weight ~ Asym/(1+exp((xmid-time)/scal)),

    start=c(Asym=30, xmid=40, scal=10),

  )

}

 

my_model <- make_model()

# Issue: when data are given in the local environment but not with a `data`

# argument, it works.

```

 

<sup>Created on 2021-08-01 by the [reprex
package](https://reprex.tidyverse.org) (v2.0.0)</sup>

 

Thanks,

 

Bill

 

P.S.  I'm pretty sure that this is also related to an issue found in the
nlraa library described at https://github.com/femiguez/nlraa/issues/4
Phillip Alday
# 
This is a bug in the way gnls is using environments to call nls() to get
starting values for glns(). You can circumvent this by doing the call to
nls() for starting values directly:

make_model <- function() {
?
? my_soybean <- Soybean
?
? form <- weight ~ SSlogis(Time, Asym, xmid, scal)
?
? gnls(form, data=my_soybean, weights=varPower(),
?????? start=coef(nls(form, my_soybean)))
?
}

Phillip
On 01/08/2021 20:43, bill at denney.ws wrote: