add butcher methods for nestedmodels

DESCRIPTION

@@ -52,6 +52,7 @@ Suggests:
     mda,
     mgcv,
     modeldata,
+    nestedmodels,
     nnet,
     parsnip (>= 0.1.6),
     pkgload,
@@ -69,6 +70,7 @@ Suggests:
     survival (>= 3.2-10),
     testthat (>= 3.0.0),
     TH.data,
+    tidyr,
     usethis (>= 1.5.0),
     xgboost (>= 1.3.2.1),
     xrf

NAMESPACE

@@ -23,6 +23,7 @@ S3method(axe_call,mixo_spls)
 S3method(axe_call,ml_model)
 S3method(axe_call,model_fit)
 S3method(axe_call,multnet)
+S3method(axe_call,nested_model_fit)
 S3method(axe_call,nnet)
 S3method(axe_call,randomForest)
 S3method(axe_call,ranger)
@@ -43,6 +44,7 @@ S3method(axe_ctrl,default)
 S3method(axe_ctrl,gam)
 S3method(axe_ctrl,ml_model)
 S3method(axe_ctrl,model_fit)
+S3method(axe_ctrl,nested_model_fit)
 S3method(axe_ctrl,randomForest)
 S3method(axe_ctrl,regbagg)
 S3method(axe_ctrl,rpart)
@@ -63,6 +65,7 @@ S3method(axe_data,mixo_pls)
 S3method(axe_data,mixo_spls)
 S3method(axe_data,ml_model)
 S3method(axe_data,model_fit)
+S3method(axe_data,nested_model_fit)
 S3method(axe_data,regbagg)
 S3method(axe_data,rpart)
 S3method(axe_data,survbagg)
@@ -85,6 +88,7 @@ S3method(axe_env,lda)
 S3method(axe_env,lm)
 S3method(axe_env,mda)
 S3method(axe_env,model_fit)
+S3method(axe_env,nested_model_fit)
 S3method(axe_env,nnet)
 S3method(axe_env,qda)
 S3method(axe_env,quosure)
@@ -132,6 +136,7 @@ S3method(axe_fitted,mixo_pls)
 S3method(axe_fitted,mixo_spls)
 S3method(axe_fitted,ml_model)
 S3method(axe_fitted,model_fit)
+S3method(axe_fitted,nested_model_fit)
 S3method(axe_fitted,nnet)
 S3method(axe_fitted,ranger)
 S3method(axe_fitted,recipe)

NEWS.md

@@ -1,8 +1,10 @@
 # butcher (development version)
 
+* Added methods for `nestedmodels::nested()` (@ashbythorpe, #256).
+
 * Updated methods for `mgcv::gam()` to also remove the `hat` and `offset` 
   components (@rdavis120, #255).
-  
+
 # butcher 0.3.2
 
 * Added butcher methods for `mixOmics::pls()`, `mixOmics::spls()`, 

R/nested_model_fit.R

@@ -0,0 +1,115 @@
+#' Axing a nested_model_fit.
+#'
+#' nested_model_fit objects are created from the \pkg{nestedmodels}
+#' package, which allows parsnip models to be fitted on nested data. Axing a
+#' nested_model_fit object involves axing all the inner model_fit objects.
+#'
+#' @inheritParams butcher
+#'
+#' @seealso [axe-model_fit]
+#'
+#' @return Axed nested_model_fit object.
+#'
+#' @examplesIf rlang::is_installed(c("parsnip", "nestedmodels"))
+#'
+#' library(nestedmodels)
+#' library(parsnip)
+#'
+#' model <- linear_reg() %>%
+#'   set_engine("lm") %>%
+#'   nested()
+#'
+#' nested_data <- tidyr::nest(example_nested_data, data = -id)
+#'
+#' fit <- fit(model, z ~ x + y + a + b, nested_data)
+#'
+#' # Reduce the model size
+#' butcher(fit)
+#'
+#' @name axe-nested_model_fit
+NULL
+
+#' Remove the call.
+#'
+#' @rdname axe-nested_model_fit
+#' @export
+axe_call.nested_model_fit <- function(x, verbose = FALSE, ...) {
+  old <- x
+  x$fit$.model_fit <- purrr::map(
+    x$fit$.model_fit,
+    axe_call,
+    verbose = FALSE,
+    ...
+  )
+
+  disabled <- attr(x$fit$.model_fit[[1]]$fit, "butcher_disabled")
+  add_butcher_attributes(x, old, disabled = disabled, verbose = verbose)
+}
+
+#' Remove controls used for training.
+#'
+#' @rdname axe-nested_model_fit
+#' @export
+axe_ctrl.nested_model_fit <- function(x, verbose = FALSE, ...) {
+  old <- x
+  x$fit$.model_fit <- purrr::map(
+    x$fit$.model_fit,
+    axe_ctrl,
+    verbose = FALSE,
+    ...
+  )
+
+  disabled <- attr(x$fit$.model_fit[[1]]$fit, "butcher_disabled")
+  add_butcher_attributes(x, old, disabled = disabled, verbose = verbose)
+}
+
+#' Remove the training data.
+#'
+#' @rdname axe-nested_model_fit
+#' @export
+axe_data.nested_model_fit <- function(x, verbose = FALSE, ...) {
+  old <- x
+  x$fit$.model_fit <- purrr::map(
+    x$fit$.model_fit,
+    axe_data,
+    verbose = FALSE,
+    ...
+  )
+
+  disabled <- attr(x$fit$.model_fit[[1]]$fit, "butcher_disabled")
+  add_butcher_attributes(x, old, disabled = disabled, verbose = verbose)
+}
+
+#' Remove environments.
+#'
+#' @rdname axe-nested_model_fit
+#' @export
+axe_env.nested_model_fit <- function(x, verbose = FALSE, ...) {
+  old <- x
+  x$fit$.model_fit <- purrr::map(
+    x$fit$.model_fit,
+    axe_env,
+    verbose = FALSE,
+    ...
+  )
+
+  disabled <- attr(x$fit$.model_fit[[1]]$fit, "butcher_disabled")
+  add_butcher_attributes(x, old, disabled = disabled, verbose = verbose)
+}
+
+#' Remove fitted values.
+#'
+#' @rdname axe-nested_model_fit
+#' @export
+axe_fitted.nested_model_fit <- function(x, verbose = FALSE, ...) {
+  old <- x
+  x$fit$.model_fit <- purrr::map(
+    x$fit$.model_fit,
+    axe_fitted,
+    verbose = FALSE,
+    ...
+  )
+
+  disabled <- attr(x$fit$.model_fit[[1]]$fit, "butcher_disabled")
+  add_butcher_attributes(x, old, disabled = disabled, verbose = verbose)
+}

tests/testthat/test-nested_model_fit.R

@@ -0,0 +1,53 @@
+test_that("nested_model_fit + axe_() works", {
+  skip_if_not_installed("parsnip")
+  skip_if_not_installed("nestedmodels")
+
+  model <- nestedmodels::nested(
+    parsnip::set_engine(parsnip::linear_reg(), "lm")
+  )
+
+  # tidyr is a dependency of nestedmodels
+  nested_data <- tidyr::nest(nestedmodels::example_nested_data, data = -id)
+
+  nm_fit <- parsnip::fit(model, z ~ x + y + a + b, nested_data)
+
+  x <- axe_call(nm_fit)
+
+  expect_equal(x$fit$.model_fit[[1]], axe_call(nm_fit$fit$.model_fit[[1]]))
+
+  x <- axe_ctrl(nm_fit)
+
+  expect_equal(x$fit$.model_fit[[1]], axe_ctrl(nm_fit$fit$.model_fit[[1]]))
+
+  x <- axe_data(nm_fit)
+
+  expect_equal(x$fit$.model_fit[[1]], axe_data(nm_fit$fit$.model_fit[[1]]))
+
+  x <- axe_env(nm_fit)
+
+  expect_equal(x$fit$.model_fit[[1]], axe_env(nm_fit$fit$.model_fit[[1]]))
+
+  x <- axe_fitted(nm_fit)
+
+  expect_equal(x$fit$.model_fit[[1]], axe_fitted(nm_fit$fit$.model_fit[[1]]))
+
+  expect_equal(
+    attr(x, "butcher_disabled"),
+    attr(x$fit$.model_fit[[5]]$fit, "butcher_disabled")
+  )
+
+  x <- butcher(nm_fit)
+
+  expect_equal(
+    attr(x, "butcher_disabled"),
+    attr(x$fit$.model_fit[[1]]$fit, "butcher_disabled")
+  )
+
+  expect_equal(x$fit$.model_fit[[1]], butcher(nm_fit$fit$.model_fit[[1]]))
+
+  # Predict
+  expect_equal(
+    predict(x, nestedmodels::example_nested_data),
+    predict(nm_fit, nestedmodels::example_nested_data)
+  )
+})