How to remove variables from an R formula

Question

I'm trying to write the following formula:

everything_time <- system.time(everything_rsf <- rfsrc(  Surv(time = PFS, event = Censoring, type = c("right")) 
                            ~.-c(Sample.ID,PatientID,UniqueID,PFS,Censoring), 
                           data = data_endogenous))

As you can see, I remove a whole bunch of variables. Yet, when I do everything_rsf$importance, I still have Sample.ID, PatientID, UniqueID in the model. I'm not sure why.

I've tried listing them out individually as well.

Can you explain what led you to believe that that syntax would successfully remove or exclude variables from the formula? — joran, Jun 23 '14 at 17:31
The model solution here shows how to remove one variable. I need to remove several. I tried listing them all out individually as well. http://stackoverflow.com/questions/5251507/how-to-succinctly-write-a-formula-with-many-variables-from-a-data-frame — user1357015, Jun 23 '14 at 17:31
Ok, you'll note that there's no use of `c()` there, so that _definitely_ won't work. — joran, Jun 23 '14 at 17:33
Before commenting on you 'listing them out individually', I'd want to see the exact syntax you used for that too. — joran, Jun 23 '14 at 17:34
Maybe instead of `data = data_endogenous` do something like `data = data_endogenous[, !(names(data_endogenous) %in% c("Sample.ID","PatientID","UniqueID",PFS,"Censoring"))]` ? — David Arenburg, Jun 23 '14 at 17:34
@joran The listing out individually went like this: ~. -Sample.ID -PatientID -UniqueID — user1357015, Jun 23 '14 at 17:48
@DavidArenburg: Yeah, I ended up doing something very similary, just cutting the matrix ahead of time by removing the columns. That gave me the stats I wanted but I'm curious how the formula would work from a programming point of view when I want to remove variables. — user1357015, Jun 23 '14 at 17:51

score 1 · Accepted Answer · answered Nov 28 '15 at 14:50

Formula won't accept list or vector arguments, a string be created.

data(iris)

fmla <- as.formula(paste("Species ~",
                   paste(c(".","Sepal.Length","Petal.Width"), collapse = " - ")))

glm(fmla, data = iris, family = binomial(link = "logit"))

Specific predictors can be defined too.

data(iris)

fmla <- as.formula(paste("Species ~", 
                   paste(grep("Width", names(iris), value = TRUE), collapse = " + ")))

glm(fmla, data = iris, family = binomial(link = "logit"))

Can combine these methods removing and adding predictors as desired, while leaving your dataset alone. Can be useful when making multiple models with the same dataset.

How to remove variables from an R formula

1 Answers1