R: Variable Ranking model automation codes to write it as a Function

Question

How can I write the list of commands below into just one Function?

For example: VariableRanking <- function(formula, variables,.....) { Insert commands........ }

#Variable Ranking Model automation
#exclusion of the variables that are not model variables
exclude <- c("~,", "+" ) # exclude target which is bound_count for Property
formula <- toString(formula)
formula

#listing the entire model formula out
variables_pre <- unlist(strsplit(formula, split = " "))
variables_pre

#keeping only the model variables
variables <-  sort(variables_pre[!variables_pre %in% exclude])
variables

#Exclude "," on the target variable 
variables[1] <- substr(variables[1], 1, nchar(variables[1])-1)
variables

#Assigning the variables into a data frame
d <- c(1:length(variables))
d
d= data.frame(d)
d
d= t(d)
d
colnames(d)=variables
d

# exclude target variable on the data frame
allvariables <- colnames(d)[-1]
allvariables
# container for models
listOfModels <- vector("list", length(allvariables))
listOfModels
# loop over variables
for (i in seq_along(allvariables)) {
  # exclude variable i
  currentvariable <- allvariables[-i]
  # programmatically assemble regression formula
  regressionFormula <- as.formula(
    paste(variables[1],"~", paste(currentvariable, collapse="+")))
  # fit model
  currentModel <- glm(formula = regressionFormula, family=binomial(link = "logit"), data=dataL_TT)
  # store model in container
  listOfModels[[i]] <- currentModel
} 
listOfModels

#List of AICs for each model 
lapply(listOfModels,function(xx) xx$aic)

#Assign X as the AIC of the full model
X <- modelTT$aic
X

# Difference of AICs of each model to the AIC of the full model
AICdifference <- lapply(listOfModels,function(xx) xx$aic - X)
AICdifference

# Naming the AIC Difference
AICdifference2 = data.frame(variables=allvariables, AICdiff=unlist(AICdifference))
AICdifference2

#Graph the Barchart of the AIC decrease of each variables and save it to pdf

pdf("Barchart.pdf",width=12,height=10)
par(mar=c(2,18,2,5))

barplot(sort(AICdifference2$AICdiff, decreasing = F), main="Variable Ranking based on AIC decrease", 
        horiz=TRUE, xlab="AIC Increase", names.arg= AICdifference2$variables[order(AICdifference2$AICdiff, decreasing = F)], 
        las=1, col= 'dodgerblue4')

dev.off()

Is it possible? because it has a lot of parameters. So basically I just need the output of the AICdifference2 data frame. And the barplot saved as pdf and pop up

Answer 1

Try this:

FOO <- function(myformula, data, fullmodel_AIC, plotname){

  exclude <- c("~,", "+" ) # exclude target which is bound_count for Property
  myformula <- toString(myformula)

  variables_pre <- unlist(strsplit(myformula, split = " "))
  variables <-  sort(variables_pre[!variables_pre %in% exclude])
  variables[1] <- substr(variables[1], 1, nchar(variables[1])-1)

  d <- t(data.frame(c(1:length(variables))))
  colnames(d)=variables

  allvariables <- colnames(d)[-1]

  listOfModels <- vector("list", length(allvariables))

  for (i in seq_along(allvariables)) {
    # exclude variable i
    currentvariable <- allvariables[-i]
    # programmatically assemble regression formula
    regressionFormula <- as.formula(
      paste(variables[1],"~", paste(currentvariable, collapse="+")))
    # fit model
    currentModel <- glm(formula = regressionFormula, family=binomial(link = "logit"), data = data)
    # store model in container
    listOfModels[[i]] <- currentModel
  } 

  AICdifference <- lapply(listOfModels,function(xx) xx$aic - fullmodel_AIC)
  AICdifference2 <- data.frame(variables=allvariables, AICdiff=unlist(AICdifference))

  pdf(paste0(plotname, ".pdf"),width=12,height=10)
  par(mar=c(2,18,2,5))

  barplot(sort(AICdifference2$AICdiff, decreasing = F), main="Variable Ranking based on AIC decrease", 
          horiz=TRUE, xlab="AIC Increase", names.arg= AICdifference2$variables[order(AICdifference2$AICdiff, decreasing = F)], 
          las=1, col= 'dodgerblue4')

  dev.off()

  return(AICdifference2)
}

You need four parameters: The myformula, the data (dataL_TT in your code), the fullmodel_AIC (modelTT$aic in your code), and a string to name your plot.

Try calling it with FOO(myformula, dataL_TT, modelTT$aic, "test") and insert your formula object for myformula.

I've changed formula to myformula because formula is a base function of the stats package, and it is generally unwise to use object names which are base functions.