Models are not found when they are elements of a list.

[giuliogcantone]

I have a model called PQP

This code:

run_specs(df = db$db1,
          y = c("y"),
          x = c("x_000"),
          model = "PQP"
          ) -> a

runs with no errors.

Now I enlist PQP in the list models. This code:

run_specs(df = db$db1,
          y = c("y"),
          x = c("x_000"),
          model = "models$PQP"
          ) -> a

gives back this error:

Error in `dplyr::mutate()`:
! Problem while computing `res = map2(.data$model, formula,
  ~do.call(.x, list(data = df, formula = .y)))`.
Caused by error in `models$PQP`:
! could not find function "models$PQP"

UPDATE:

TimTeaFan provided a debug:

get_model <- function(x) {
  x_str <- str2lang(x)
  if (is.name(x_str)) {
    return(x)
    } else if (is.call(x_str)) {
    eval(x_str)
  }
}

Then specr:::run_spec is corrected like this:

specs %>%
  dplyr::mutate(formula = pmap(.,
                               specr:::create_formula)
  ) %>% tidyr::unnest(formula) %>% 
  dplyr::mutate(res = map2(model,
                           formula,
                           ~ do.call(get_model(.x), list(data = df, formula = .y))))

[giuliogcantone]

I am trying to debug this, and I located the source of the bug in the internal code of run_spec

dplyr::mutate(res = map2(.data$model,
                           formula,
                           ~do.call(.x,
                                    list(data = df,
                                         formula = .y))))

[lmullany]

.x must be a character string naming the function to be used, or the function it self.. for example do.call("rbind", ...) or do.call(rbind, ...).

However, if you pass the character string "models$PQP" in run_specs(), then .x will be "models$PQP" (which is NOT a function or a character string naming the function)..

[giuliogcantone]

.x must be a character string naming the function to be used, or the function it self.. for example do.call("rbind", ...) or do.call(rbind, ...).

However, if you pass the character string "models$PQP" in run_specs(), then .x will be "models$PQP" (which is NOT a function or a character string naming the function)..

My practical problem is that I want to process 32 model specifications in feglm(). Since these have a | in the formula, I have to code separately each of them as a model function, and then process them in the models param. I could set a env with 32 functions objects and that could work, but it is very not elegant. So I wanted to process them as elements of a list. If they are saved in the env, for example, they will keep popping while typing, which is ugly.

Maybe a solution is kinda... make a package p_adhoc of these 32 functions and call them with `p_adhoc::function"? I've never made a package tho.

Have you suggestions to not clog env with functions? I think that this is relevant to make advance this package for fixed effect models. EG, one may test more than 100 ad hoc fixed effects models.

[giuliogcantone]

UPDATE:

TimTeaFan provided a debug:

get_model <- function(x) {
  x_str <- str2lang(x)
  if (is.name(x_str)) {
    return(x)
    } else if (is.call(x_str)) {
    eval(x_str)
  }
}

Then specr:::run_spec is corrected like this:

specs %>%
  dplyr::mutate(formula = pmap(.,
                               specr:::create_formula)
  ) %>% tidyr::unnest(formula) %>% 
  dplyr::mutate(res = map2(model,
                           formula,
                           ~ do.call(get_model(.x), list(data = df, formula = .y))))

[masurp]

Hey guys, sorry for not chipping in earlier. I am currently on parental leave, hence have only limited time. Thanks to all for the contribution/discussion. I have been working on a new version of specr for a while (including several other aspects) and this will be interesting to add. Will get into it hopefully very soon!

Best, Philipp