How do I fill the subset argument with a variable?

Question

I'd like to use a variable for the subset argument so I can put it into a function

     formula <- paste0(response_name, 
                         " ~ .")
     
     if (subset_filter != ""){
       subset_filter <- "G3 < 10"
       
       model <- 
         lm(as.formula(formula),
            subset = subset_filter,
            data = train_dataset)    
       
     } else {
      model <- 
         lm(as.formula(formula),
            data = train_dataset)
      
     }

My dataset is this -

student_performance <-
  read_csv("https://raw.githubusercontent.com/UBC-MDS/ellognea-smwatts-student-performance/master/data/student-math-perf.csv") %>% 
  as_tibble()

And my response variable is G3 and I split the sets with this code

split_sets <- function(dataset, 
                       response_name,
                       output_set_type){

  set.seed(1)
  training.samples <- createDataPartition(as_vector(dataset[response_name]), 
                                          p = 0.8,
                                          list = F)
  
  train.data <- suppressWarnings(dataset[training.samples, ])
  test.data <- suppressWarnings(dataset[-training.samples, ])  
  
  l <- list()
  
  l[["train.data"]] <-
    train.data
  
  l[["test.data"]] <-
    test.data
  
  ifelse(output_set_type == "train", 
         return(as_tibble(l$train.data)), 
         return(as_tibble(l$test.data)))

}

I'd like to make it so that I can submit values into the subset filter argument and then use them

Answer 1

If we need to pass a string, then we could parse and eval uate

library(caret)
library(readr)    
create_model <- function(data, response_name, subset_filter) {

     
     formula <- paste0(response_name, " ~ .")    
     
     if (subset_filter != ""){             
       
       model <- 
         lm(as.formula(formula),
            subset = eval(parse(text = subset_filter), envir = data),
            data = data)    
       
     } else {
      model <- 
         lm(as.formula(formula),
            data = data)
      
     }
     
     model$call <- as.formula(formula)
     return(model)

}

-apply the function on the data

create_model(train_dat, "G3", "G3 < 10" )
#Call:
#G3 ~ .

#Coefficients:
#     (Intercept)          schoolMS              sexM               age          addressU        famsizeLE3  
#        -4.42602           1.18145           0.15315          -0.16790           1.11708          -0.08173  
#        PstatusT              Medu              Fedu        Mjobhealth         Mjobother      Mjobservices  
#         1.32870           1.00518          -0.62716          -1.98356          -1.31388          -0.94443  
#     Mjobteacher        Fjobhealth         Fjobother      Fjobservices       Fjobteacher        reasonhome  
#        -1.28718           0.03242           0.02968           0.32962          -1.53201          -2.10665  
#     reasonother  reasonreputation    guardianmother     guardianother        traveltime         studytime  
#        -0.51770           0.22395          -0.29893           1.85975          -0.39072          -1.56920  
#        failures      schoolsupyes         famsupyes           paidyes     activitiesyes        nurseryyes  
#        -0.17344           2.35607           0.35207           0.29857          -0.91373           0.09838  
#       higheryes       internetyes       romanticyes            famrel          freetime             goout  
#         1.06065          -0.58727           0.09469           0.69217           0.25081           0.14379  
#            Dalc              Walc            health          absences                G1                G2  
#       -1.39164           0.60450           0.86492           0.12033           0.11660           0.78624  

---

Here, the 'train_data' is created from

split_sets <- function(dataset, 
                       response_name,
                       output_set_type){

  set.seed(1)
  training.samples <- createDataPartition(as_vector(dataset[response_name]), 
                                          p = 0.8,
                                          list = F)
  
  train.data <- suppressWarnings(dataset[training.samples, ])
  test.data <- suppressWarnings(dataset[-training.samples, ])  
  
  l <- list()
  
  l[["train.data"]] <-
    train.data
  
  l[["test.data"]] <-
    test.data
  
   out <- if(output_set_type == "train") {
   
        as_tibble(l$train.data)
        
        } else {
        as_tibble(l$test.data)
        }
        
  return(out)

}


train_dat <- split_sets(student_performance, "G3", "train")