cross_validate {origami}

Main Cross-Validation Function

Description

Applies cv_fun to the folds using future_lapply and combines the results across folds using combine_results.

Usage

cross_validate(
  cv_fun,
  folds,
  ...,
  use_future = TRUE,
  .combine = TRUE,
  .combine_control = list(),
  .old_results = NULL
)

Arguments

cv_fun

A function that takes a 'fold' as it's first argument and returns a list of results from that fold. NOTE: the use of an argument named 'X' is specifically disallowed in any input function for compliance with the functions future_lapply and lapply.

folds

A list of folds to loop over generated using make_folds.

...

Other arguments passed to cvfun.

use_future

A logical option for whether to run the main loop of cross-validation with future_lapply or with lapply.

.combine

A logical indicating if combine_results should be called.

.combine_control

A list of arguments to combine_results.

.old_results

A list containing the returned result from a previous call to this function. Will be combined with the current results. This is useful for adding additional CV folds to a results object.

Value

A list of results, combined across folds.

Examples

###############################################################################
# This example explains how to use the cross_validate function naively.
###############################################################################
data(mtcars)

# resubstitution MSE
r <- lm(mpg ~ ., data = mtcars)
mean(resid(r)^2)

# function to calculate cross-validated squared error
cv_lm <- function(fold, data, reg_form) {
  # get name and index of outcome variable from regression formula
  out_var <- as.character(unlist(stringr::str_split(reg_form, " "))[1])
  out_var_ind <- as.numeric(which(colnames(data) == out_var))

  # split up data into training and validation sets
  train_data <- training(data)
  valid_data <- validation(data)

  # fit linear model on training set and predict on validation set
  mod <- lm(as.formula(reg_form), data = train_data)
  preds <- predict(mod, newdata = valid_data)

  # capture results to be returned as output
  out <- list(
    coef = data.frame(t(coef(mod))),
    SE = ((preds - valid_data[, out_var_ind])^2)
  )
  return(out)
}

# replicate the resubstitution estimate
resub <- make_folds(mtcars, fold_fun = folds_resubstitution)[[1]]
resub_results <- cv_lm(fold = resub, data = mtcars, reg_form = "mpg ~ .")
mean(resub_results$SE)

# cross-validated estimate
folds <- make_folds(mtcars)
cv_results <- cross_validate(
  cv_fun = cv_lm, folds = folds, data = mtcars,
  reg_form = "mpg ~ ."
)
mean(cv_results$SE)
###############################################################################
# This example explains how to use the cross_validate function with
# parallelization using the framework of the future package.
###############################################################################

suppressMessages(library(data.table))
library(future)
data(mtcars)
set.seed(1)

# make a lot of folds
folds <- make_folds(mtcars, fold_fun = folds_bootstrap, V = 1000)

# function to calculate cross-validated squared error for linear regression
cv_lm <- function(fold, data, reg_form) {
  # get name and index of outcome variable from regression formula
  out_var <- as.character(unlist(str_split(reg_form, " "))[1])
  out_var_ind <- as.numeric(which(colnames(data) == out_var))

  # split up data into training and validation sets
  train_data <- training(data)
  valid_data <- validation(data)

  # fit linear model on training set and predict on validation set
  mod <- lm(as.formula(reg_form), data = train_data)
  preds <- predict(mod, newdata = valid_data)

  # capture results to be returned as output
  out <- list(
    coef = data.frame(t(coef(mod))),
    SE = ((preds - valid_data[, out_var_ind])^2)
  )
  return(out)
}

plan(sequential)
time_seq <- system.time({
  results_seq <- cross_validate(
    cv_fun = cv_lm, folds = folds, data = mtcars,
    reg_form = "mpg ~ ."
  )
})

plan(multicore)
time_mc <- system.time({
  results_mc <- cross_validate(
    cv_fun = cv_lm, folds = folds, data = mtcars,
    reg_form = "mpg ~ ."
  )
})

if (availableCores() > 1) {
  time_mc["elapsed"] < 1.2 * time_seq["elapsed"]
}