Fit a stats::glm without carying back large structures.

Source: R/clean_fit.R

Please see https://win-vector.com/2014/05/30/trimming-the-fat-from-glm-models-in-r/ for discussion.

clean_fit_glm(
  outcome,
  variables,
  data,
  ...,
  family,
  intercept = TRUE,
  outcome_target = NULL,
  outcome_comparator = "==",
  weights = NULL,
  env = baseenv()
)

Arguments

outcome

character, name of outcome column.
variables

character, names of varaible columns.
data

data.frame, training data.
...

not used, force later arguments to be used by name
family

passed to stats::glm()
intercept

logical, if TRUE allow an intercept term.
outcome_target

scalar, if not NULL write outcome==outcome_target in formula.
outcome_comparator

one of "==", "!=", ">=", "<=", ">", "<", only use of outcome_target is not NULL.
weights

passed to stats::glm()
env

environment to work in.

Value

list(model=model, summary=summary)

Examples


mk_data_example <- function(k) {
  data.frame(
    x1 = rep(c("a", "a", "b", "b"), k),
    x2 = rep(c(0, 0, 0, 1), k),
    y = rep(1:4, k),
    yC = rep(c(FALSE, TRUE, TRUE, TRUE), k),
    stringsAsFactors = FALSE)
}

res_glm <- clean_fit_glm("yC", c("x1", "x2"),
                         mk_data_example(1),
                         family = binomial)
length(serialize(res_glm$model, NULL))

#> [1] 33777

res_glm <- clean_fit_glm("yC", c("x1", "x2"),
                         mk_data_example(10000),
                         family = binomial)
length(serialize(res_glm$model, NULL))

#> [1] 33777

predict(res_glm$model,
        newdata = mk_data_example(1),
        type = "response")

#>   1   2   3   4 
#> 0.5 0.5 1.0 1.0