Is there any function in R, whose return is identical to the return of excel’s SEERRO() function?

Question

I am trying to generate a general function, which runs several regression models. The problem is that when a model gives error, the main function stops. I would like to use a parole that in case of error, skip to the next step:

Example: when trying to run a model nls(), presents this message described below, as this function is one of which I am testing within a general function, it locks the function at this point.

an example, when I want to load some package, I usually do so:

if(!require(nome do pacote)){install.packages("nome do pacote")}

in my intention is like this: if(there is no such package){ install the package} so here I inform that "!" would represent the "error".

assuming I create a function:

melhor.ajuste<-function(df,x,y){
n1<-nls(y~x+c/e*x^d,start=....,data=df)
n2<-nls(y~x-b*e^d/log(x),start=....,data=df)
n3<-nls(y~x+a*b/c^x,start=....,data=df)
list(modelo1=n1,modelo2=n2,modelo3=n3)}

if for example the function N2 shows any error, as an example:

stop(simpleError("Error in nlsModel(formula, mf, start, wts) :\n matriz gradiente singular com estimativas de parâmetros iniciais"))

my function does not present me any value, because there was a mistake!

what I wish was something like:

if(!nls(y~x+c/e*x^d,start=....,data=df)){n1=0}else{n1<-nls(y~x+c/e*x^d,start=....,data=df)}
if(!nls(y~x+a*b/c^x,start=....,data=df)){n2=0}else{n1<-nls(y~x+a*b/c^x,start=....,data=df)}
if(!nls(y~x+a*b/c^x,start=....,data=df)){n3=0}else{n1<-nls(y~x+a*b/c^x,start=....,data=df)}

someone knows somehow I do it! excel does, by the SEERRO() function, where I go (if I make a mistake; do this), in my case it would be, se(seerro(nls(y~x+a*b/c^x,start=....,data=df);0)==0;n1=0;n1=nls(y~x+a*b/c^x,start=....,data=df)), or would be a parole that identifies error, within a conditional.

Making it clear, that at no time, my main objective is to find out what the mistake is and try to fix it, because I will apply the function in more than 50 models, knowing that some do not fit even, because it is a singular matrix, for bursting the minFactor, and something like that. my goal, is to find a way that if the function presents error it concatene 0 to the object, and pass to the next model.

and at the end provide me the list of all models.

Answer 1

Perhaps this artificial example, as is said in one of the examples of help("nls") in which you are inspired, you can give a helping hand. It shows how you can try several templates and continue even if there are errors in any of them. This is done using the function tryCatch error, a list of two members.

Data at the end.

regr <- "a + b*x"
resp <- grep("^y", names(dados), value = TRUE)

fit_list <- lapply(resp, function(y){
  fmla <- paste(y, regr, sep = "~")
  fmla <- as.formula(fmla)
  tryCatch(
    nls(fmla, start = list(a = 0.12345, b = 0.54321)),
    error = function(e) e
  )
})

err <- sapply(fit_list, inherits, "error")

Now we can see what the error structure is.

str(fit_list[err][[1]])
#List of 2
# $ message: chr "number of iterations exceeded maximum of 50"
# $ call   : language nls(fmla, start = list(a = 0.12345, b = 0.54321))
# - attr(*, "class")= chr [1:3] "simpleError" "error" "condition"

And get the member "message" for each of the list errors fit_list[err].

lapply(fit_list[err], "[[", "message")
#[[1]]
#[1] "number of iterations exceeded maximum of 50"
#
#[[2]]
#[1] "number of iterations exceeded maximum of 50"

As for the models that went well, one can extract the information available also with cycles lapply.

lapply(fit_list[!err], coef)
lapply(fit_list[!err], AIC)
lapply(fit_list[!err], summary)

---

With a for, the code has some changes but the call to tryCatch is the same.

The change is how to store the results. The best way is to create a list beforehand and assign the output of trCatch to each member of that list.

fit_list_for_loop <- vector("list", length = length(resp))

for(i in seq_along(resp)){
  fmla <- paste(resp[i], regr, sep = "~")
  fmla <- as.formula(fmla)
  fit_list_for_loop[[i]] <- tryCatch(
    nls(fmla, start = list(a = 0.12345, b = 0.54321)),
    error = function(e) e
  )
}

all.equal(fit_list, fit_list_for_loop)
#[1] TRUE

Test data

set.seed(2021)
x <- 1:10
y1 <- 2*x + 3                              # perfect fit
y1eps <- y1 + rnorm(length(y1), sd = 0.01) # added noise
y2 <- 3*x - 2                              # perfect fit
y2eps <- y2 + rnorm(length(y2), sd = 0.01) # added noise
dados <- data.frame(x, y1, y1eps, y2, y2eps)