问题
I am trying to use quosures to pass along variable names within a custom function for data processing and use in a formula, but my use of quosures in the formula is not correct. Is there a better way to unquote arguments within a formula?
library(dplyr)
library(broom)
library(purrr)
library(tidyr)
foo <- function(mydata, dv, iv, group_var) {
dv = enquo(dv)
iv = enquo(iv)
group_var = enquo(group_var)
mydata <- mydata %>%
group_by(!!group_var) %>%
nest()
mydata %>%
mutate(model = map(data,
~summary(lm(formula(substitute(dv ~ iv)), data = .))
)) %>%
unnest(model %>% map(tidy))
}
foo(mydata=mtcars, dv=mpg, iv=wt, group_var=cyl)
My code produces "Error in mutate_impl(.data, dots) : Evaluation error: object is not a matrix."
This is a working version of code I am trying to make into a function:
mtcars %>%
group_by(cyl) %>%
nest() %>%
mutate(model = map(data, ~summary(lm(mpg ~ wt, data = .)))) %>%
unnest(model %>% map(tidy))
回答1:
You need to use base R nonstandard evaluation with functions like lm
which are not "in the tidyverse" so to speak.
So you could change things to:
foo <- function(mydata, dv, iv, group_var) {
flma <- as.formula(paste(substitute(dv), "~", substitute(iv)))
group_var = enquo(group_var)
mydata <- mydata %>%
group_by(!!group_var) %>%
nest()
mydata %>%
mutate(model = map(data, ~summary(lm(flma, data = .)))) %>%
unnest(model %>% map(tidy))
}
foo(mtcars, mpg, wt, cyl)
That's fine if you know you are only doing simple regression. For more flexibility just pass the formula directly, as in:
foo2 <- function(mydata, flma, group_var) {
group_var = enquo(group_var)
mydata <- mydata %>%
group_by(!!group_var) %>%
nest()
mydata %>%
mutate(model = map(data, ~summary(lm(flma, data = .)))) %>%
unnest(model %>% map(tidy))
}
foo(mtcars, mpg ~ wt, cyl)
来源:https://stackoverflow.com/questions/50053035/using-quosures-within-formula-inside-an-anonymous-function