Package 'butcher'

Title: Model Butcher
Description: Provides a set of S3 generics to axe components of fitted model objects and help reduce the size of model objects saved to disk.
Authors: Joyce Cahoon [aut] , Davis Vaughan [aut], Max Kuhn [aut], Alex Hayes [aut], Julia Silge [cre, aut] , Posit, PBC [cph, fnd]
Maintainer: Julia Silge <[email protected]>
License: MIT + file LICENSE
Version: 0.3.4.9000
Built: 2024-10-25 05:19:31 UTC
Source: https://github.com/tidymodels/butcher

Help Index

Axe a call.

Description

Replace the call object attached to modeling objects with a placeholder.

Usage

axe_call(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Model object without call attribute.

Methods

See the following help topics for more details about individual methods:

butcher

Axe controls.

Description

Remove the controls from training attached to modeling objects.

Usage

axe_ctrl(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Model object without control tuning parameters from training.

Methods

See the following help topics for more details about individual methods:

butcher

Axe data.

Description

Remove the training data attached to modeling objects.

Usage

axe_data(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Model object without the training data

Methods

See the following help topics for more details about individual methods:

butcher

Axe an environment.

Description

Remove the environment(s) attached to modeling objects as they are not required in the downstream analysis pipeline. If found, the environment is replaced with rlang::base_env().

Usage

axe_env(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Model object with empty environments.

Methods

See the following help topics for more details about individual methods:

butcher

Axe fitted values.

Description

Remove the fitted values attached to modeling objects.

Usage

axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Model object without the fitted values.

Methods

See the following help topics for more details about individual methods:

butcher

Axing a bart model.

Description

Axing a bart model.

Usage

## S3 method for class 'bart'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'bart'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed bart object.

Examples

library(dbarts)
x <- dbarts::bart(mtcars[,2:5], mtcars[,1], verbose = FALSE, keeptrees = TRUE)
res <- butcher(x, verbose = TRUE)

Axing a C5.0.

Description

C5.0 objects are created from the C50 package, which provides an interface to the C5.0 classification model. The models that can be generated include basic tree-based models as well as rule-based models.

Usage

## S3 method for class 'C5.0'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'C5.0'
axe_ctrl(x, verbose = FALSE, ...)

## S3 method for class 'C5.0'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed C5.0 object.

Examples

# Load libraries
library(parsnip)
library(rsample)
library(rpart)

# Load data
set.seed(1234)
split <- initial_split(kyphosis, prop = 9/10)
spine_train <- training(split)

# Create model and fit
c5_fit <- decision_tree(mode = "classification") %>%
  set_engine("C5.0") %>%
  fit(Kyphosis ~ ., data = spine_train)

out <- butcher(c5_fit, verbose = TRUE)

# Try another model from parsnip
c5_fit2 <- boost_tree(mode = "classification", trees = 100) %>%
  set_engine("C5.0") %>%
  fit(Kyphosis ~ ., data = spine_train)
out <- butcher(c5_fit2, verbose = TRUE)

# Create model object from original library
library(C50)
library(modeldata)
data(mlc_churn)
c5_fit3 <- C5.0(x = mlc_churn[, -20], y = mlc_churn$churn)
out <- butcher(c5_fit3, verbose = TRUE)

Axing a coxph.

Description

Axing a coxph.

Usage

## S3 method for class 'coxph'
axe_env(x, verbose = FALSE, ...)

## S3 method for class 'coxph'
axe_data(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Details

The survival::coxph() model is unique in how it uses environments in its components, and butchering such an object can behave in surprising ways in any environment other than the global environment (such as when wrapped in a function). We do not recommend that you use butcher() with a coxph object anywhere other than the global environment.

Do this: 

my_coxph_func <- function(df) {
    coxph(Surv(time, status) ~ x + strata(covar), df)
}
## in global environment only:
butcher(my_coxph_func(df))

Do not do this: 

my_coxph_func <- function(df) {
    res <- coxph(Surv(time, status) ~ x + strata(covar), df)
    ## no:
    butcher(res)
}

## will not work correctly:
my_coxph_func(df)

Value

Axed coxph object.

Examples

library(survival)

example_data <-
  tibble::tibble(
    time = rpois(1000, 2) + 1,
    status = rbinom(1000, 1, .5),
    x = rpois(1000, .5),
    covar = rbinom(1000, 1, .5)
  )

example_data

make_big_model <- function() {
  boop <- runif(1e6)
  coxph(Surv(time, status) ~ x + strata(covar), example_data)
}

res <- make_big_model()

weigh(res)
weigh(butcher(res))

Axing an earth object.

Description

earth objects are created from the earth package, which is leveraged to do multivariate adaptive regression splines.

Usage

## S3 method for class 'earth'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'earth'
axe_data(x, verbose = FALSE, ...)

## S3 method for class 'earth'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed earth object.

Examples

# Load libraries
library(parsnip)

# Create model and fit
earth_fit <- mars(mode = "regression") %>%
  set_engine("earth") %>%
  fit(Volume ~ ., data = trees)

out <- butcher(earth_fit, verbose = TRUE)

# Another earth model object
suppressWarnings(suppressMessages(library(earth)))
earth_mod <- earth(Volume ~ ., data = trees)
out <- butcher(earth_mod, verbose = TRUE)

Axing an elnet.

Description

elnet objects are created from the glmnet package, leveraged to fit generalized linear models via penalized maximum likelihood.

Usage

## S3 method for class 'elnet'
axe_call(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed model object.

Examples

# Load libraries
library(parsnip)
library(rsample)

# Load data
split <- initial_split(mtcars, prop = 9/10)
car_train <- training(split)

# Create model and fit
elnet_fit <- linear_reg(mixture = 0, penalty = 0.1) %>%
  set_engine("glmnet") %>%
  fit_xy(x = car_train[, 2:11], y = car_train[, 1, drop = FALSE])

out <- butcher(elnet_fit, verbose = TRUE)

Axing an flexsurvreg.

Description

flexsurvreg objects are created from the flexsurv package. They differ from survreg in that the fitted models are not limited to certain parametric distributions. Users can define their own distribution, or leverage distributions like the generalized gamma, generalized F, and the Royston-Parmar spline model.

Usage

## S3 method for class 'flexsurvreg'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'flexsurvreg'
axe_env(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed flexsurvreg object.

Examples

# Load libraries
library(parsnip)
library(flexsurv)

# Create model and fit
flexsurvreg_fit <- surv_reg(mode = "regression", dist = "gengamma") %>%
  set_engine("flexsurv") %>%
  fit(Surv(Tstart, Tstop, status) ~ trans, data = bosms3)

out <- butcher(flexsurvreg_fit, verbose = TRUE)

# Another flexsurvreg model object
wrapped_flexsurvreg <- function() {
  some_junk_in_environment <- runif(1e6)
  fit <- flexsurvreg(Surv(futime, fustat) ~ 1,
                     data = ovarian, dist = "weibull")
  return(fit)
}

out <- butcher(wrapped_flexsurvreg(), verbose = TRUE)

Axing formulas.

Description

formulas might capture an environment from the modeling development process that carries objects that will not be used for any post- estimation activities.

Usage

## S3 method for class 'formula'
axe_env(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed formula object.

Examples

wrapped_formula <- function() {
  some_junk_in_environment <- runif(1e6)
  ex <- as.formula(paste("y ~", paste(LETTERS, collapse = "+")))
  return(ex)
}

lobstr::obj_size(wrapped_formula())
lobstr::obj_size(butcher(wrapped_formula()))

wrapped_quosure <- function() {
  some_junk_in_environment <- runif(1e6)
  out <- rlang::quo(x)
  return(out)
}
lobstr::obj_size(wrapped_quosure())
lobstr::obj_size(butcher(wrapped_quosure))

Axing functions.

Description

Functions stored in model objects often have heavy environments and bytecode attached. To avoid breaking any post-estimation functions on the model object, the butchered_function class is not appended.

Usage

## S3 method for class ''function''
axe_env(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed function.

Examples

# Load libraries
library(caret)

data(iris)
train_data <- iris[, 1:4]
train_classes <- iris[, 5]

train_fit <- train(train_data, train_classes,
                   method = "knn",
                   preProcess = c("center", "scale"),
                   tuneLength = 10,
                   trControl = trainControl(method = "cv"))

out <- axe_env(train_fit$modelInfo$prob, verbose = TRUE)
out <- axe_env(train_fit$modelInfo$levels, verbose = TRUE)
out <- axe_env(train_fit$modelInfo$predict, verbose = TRUE)

Axing a gam.

Description

gam objects are created from the mgcv package.

Usage

## S3 method for class 'gam'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'gam'
axe_ctrl(x, verbose = FALSE, ...)

## S3 method for class 'gam'
axe_data(x, verbose = FALSE, ...)

## S3 method for class 'gam'
axe_env(x, verbose = FALSE, ...)

## S3 method for class 'gam'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed gam object.

Examples

cars_gam <- mgcv::gam(mpg ~ s(disp, k = 3) + s(wt), data = mtcars)
cleaned_gam <- butcher(cars_gam, verbose = TRUE)

Axing a gausspr.

Description

gausspr objects are created from kernlab package, which provides a means to do classification, regression, clustering, novelty detection, quantile regression and dimensionality reduction. Since fitted model objects from kernlab are S4, the butcher_gausspr class is not appended.

Usage

## S3 method for class 'gausspr'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'gausspr'
axe_data(x, verbose = FALSE, ...)

## S3 method for class 'gausspr'
axe_env(x, verbose = FALSE, ...)

## S3 method for class 'gausspr'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed gausspr object.

Examples

library(kernlab)

test <- gausspr(Species ~ ., data = iris, var = 2)

out <- butcher(test, verbose = TRUE)

# Example with simulated regression data
x <- seq(-20, 20, 0.1)
y <- sin(x)/x + rnorm(401, sd = 0.03)
test2 <- gausspr(x, y)
out <- butcher(test2, verbose = TRUE)

Axing a glm.

Description

glm objects are created from the base stats package.

Usage

## S3 method for class 'glm'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'glm'
axe_data(x, verbose = FALSE, ...)

## S3 method for class 'glm'
axe_env(x, verbose = FALSE, ...)

## S3 method for class 'glm'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed glm object.

Examples

cars_glm <- glm(mpg ~ ., data = mtcars)
cleaned_glm <- butcher(cars_glm, verbose = TRUE)

Axing a glmnet.

Description

glmnet objects are created from the glmnet package, leveraged to fit generalized linear models via penalized maximum likelihood.

Usage

## S3 method for class 'glmnet'
axe_call(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed glmnet object.

Examples

library(parsnip)

# Wrap a parsnip glmnet model
wrapped_parsnip_glmnet <- function() {
  some_junk_in_environment <- runif(1e6)
  model <- logistic_reg(penalty = 10, mixture = 0.1) %>%
    set_engine("glmnet") %>%
    fit(as.factor(vs) ~ ., data = mtcars)
  return(model$fit)
}

out <- butcher(wrapped_parsnip_glmnet(), verbose = TRUE)

Axing a bagged tree.

Description

⁠*_bagg⁠ objects are created from the ipred package, which is used for bagging classification, regression and survival trees.

Usage

## S3 method for class 'regbagg'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'classbagg'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'survbagg'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'regbagg'
axe_ctrl(x, verbose = FALSE, ...)

## S3 method for class 'classbagg'
axe_ctrl(x, verbose = FALSE, ...)

## S3 method for class 'survbagg'
axe_ctrl(x, verbose = FALSE, ...)

## S3 method for class 'regbagg'
axe_data(x, verbose = FALSE, ...)

## S3 method for class 'classbagg'
axe_data(x, verbose = FALSE, ...)

## S3 method for class 'survbagg'
axe_data(x, verbose = FALSE, ...)

## S3 method for class 'regbagg'
axe_env(x, verbose = FALSE, ...)

## S3 method for class 'classbagg'
axe_env(x, verbose = FALSE, ...)

## S3 method for class 'survbagg'
axe_env(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed ⁠*_bagg⁠ object.

Examples

library(ipred)

fit_mod <- function() {
  boop <- runif(1e6)
  bagging(y ~ x, data.frame(y = rnorm(1e4), x = rnorm(1e4)))
}

mod_fit <- fit_mod()
mod_res <- butcher(mod_fit)

weigh(mod_fit)
weigh(mod_res)

Axing an kknn.

Description

kknn objects are created from the kknn package, which is utilized to do weighted k-Nearest Neighbors for classification, regression and clustering.

Usage

## S3 method for class 'kknn'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'kknn'
axe_env(x, verbose = FALSE, ...)

## S3 method for class 'kknn'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed kknn object.

Examples

# Load libraries
library(parsnip)
library(rsample)
library(rpart)
library(kknn)

# Load data
set.seed(1234)
split <- initial_split(kyphosis, prop = 9/10)
spine_train <- training(split)

# Create model and fit
kknn_fit <- nearest_neighbor(mode = "classification",
                             neighbors = 3,
                             weight_func = "gaussian",
                             dist_power = 2) %>%
  set_engine("kknn") %>%
  fit(Kyphosis ~ ., data = spine_train)

out <- butcher(kknn_fit, verbose = TRUE)


# Another kknn model object
m <- dim(iris)[1]
val <- sample(1:m,
              size = round(m/3),
              replace = FALSE,
              prob = rep(1/m, m))
iris.learn <- iris[-val,]
iris.valid <- iris[val,]
kknn_fit <- kknn(Species ~ .,
                 iris.learn,
                 iris.valid,
                 distance = 1,
                 kernel = "triangular")
out <- butcher(kknn_fit, verbose = TRUE)

Axing a KMeansCluster.

Description

Axing a KMeansCluster.

Usage

## S3 method for class 'KMeansCluster'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'KMeansCluster'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed KMeansCluster object.

Examples

library(ClusterR)
data(dietary_survey_IBS)
dat <- scale(dietary_survey_IBS[, -ncol(dietary_survey_IBS)])
km <- KMeans_rcpp(dat, clusters = 2, num_init = 5)
out <- butcher(km, verbose = TRUE)

Axing a kproto.

Description

Axing a kproto.

Usage

## S3 method for class 'kproto'
axe_data(x, verbose = FALSE, ...)

## S3 method for class 'kproto'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed kproto object.

Examples

library(clustMixType)

kproto_fit <- kproto(
  ToothGrowth,
  k = 2,
  lambda = lambdaest(ToothGrowth),
  verbose = FALSE
)

out <- butcher(kproto_fit, verbose = TRUE)

Axing a ksvm object.

Description

ksvm objects are created from kernlab package, which provides a means to do classification, regression, clustering, novelty detection, quantile regression and dimensionality reduction. Since fitted model objects from kernlab are S4, the butcher_ksvm class is not appended.

Usage

## S3 method for class 'ksvm'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'ksvm'
axe_data(x, verbose = FALSE, ...)

## S3 method for class 'ksvm'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed ksvm object.

Examples

# Load libraries
library(parsnip)
library(kernlab)

# Load data
data(spam)

# Create model and fit
ksvm_class <- svm_poly(mode = "classification") %>%
  set_engine("kernlab") %>%
  fit(type ~ ., data = spam)

out <- butcher(ksvm_class, verbose = TRUE)

Axing an lm.

Description

lm objects are created from the base stats package.

Usage

## S3 method for class 'lm'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'lm'
axe_env(x, verbose = FALSE, ...)

## S3 method for class 'lm'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed lm object.

Examples

# Load libraries
library(parsnip)
library(rsample)

# Load data
split <- initial_split(mtcars, prop = 9/10)
car_train <- training(split)

# Create model and fit
lm_fit <- linear_reg() %>%
  set_engine("lm") %>%
  fit(mpg ~ ., data = car_train)

out <- butcher(lm_fit, verbose = TRUE)

# Another lm object
wrapped_lm <- function() {
  some_junk_in_environment <- runif(1e6)
  fit <- lm(mpg ~ ., data = mtcars)
  return(fit)
}

# Remove junk
cleaned_lm <- axe_env(wrapped_lm(), verbose = TRUE)

# Check size
lobstr::obj_size(cleaned_lm)

# Compare environment in terms component
lobstr::obj_size(attr(wrapped_lm()$terms, ".Environment"))
lobstr::obj_size(attr(cleaned_lm$terms, ".Environment"))

Axing a MASS discriminant analysis object.

Description

lda and qda objects are created from the MASS package, leveraged to carry out linear discriminant analysis and quadratic discriminant analysis, respectively.

Usage

## S3 method for class 'lda'
axe_env(x, verbose = FALSE, ...)

## S3 method for class 'qda'
axe_env(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed lda or qda object.

Examples

library(MASS)

fit_da <- function(fit_fn) {
  boop <- runif(1e6)
  fit_fn(y ~ x, data.frame(y = rep(letters[1:4], 10000), x = rnorm(40000)))
}

lda_fit <- fit_da(lda)
qda_fit <- fit_da(qda)

lda_fit_b <- butcher(lda_fit)
qda_fit_b <- butcher(qda_fit)

weigh(lda_fit)
weigh(lda_fit_b)

weigh(qda_fit)
weigh(qda_fit_b)

Axing a mda.

Description

mda and fda objects are created from the mda package, leveraged to carry out mixture discriminant analysis and flexible discriminat analysis, respectively.

Usage

## S3 method for class 'mda'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'fda'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'mda'
axe_env(x, verbose = FALSE, ...)

## S3 method for class 'fda'
axe_env(x, verbose = FALSE, ...)

## S3 method for class 'mda'
axe_fitted(x, verbose = FALSE, ...)

## S3 method for class 'fda'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed mda object.

Examples

library(mda)

mtcars$cyl <- as.factor(mtcars$cyl)

fit <- mda(cyl ~ ., data = mtcars)
out <- butcher(fit, verbose = TRUE)

fit2 <- fda(cyl ~ ., data = mtcars)
out2 <- butcher(fit2, verbose = TRUE)

# Another mda object
data(glass)
wrapped_mda <- function(fit_fn) {
  some_junk_in_environment <- runif(1e6)
  fit <- fit_fn(Type ~ ., data = glass)
  return(fit)
}

lobstr::obj_size(wrapped_mda(mda))
lobstr::obj_size(butcher(wrapped_mda(mda)))

lobstr::obj_size(wrapped_mda(fda))
lobstr::obj_size(butcher(wrapped_mda(fda)))

Axing an model_fit.

Description

model_fit objects are created from the parsnip package.

Usage

## S3 method for class 'model_fit'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'model_fit'
axe_ctrl(x, verbose = FALSE, ...)

## S3 method for class 'model_fit'
axe_data(x, verbose = FALSE, ...)

## S3 method for class 'model_fit'
axe_env(x, verbose = FALSE, ...)

## S3 method for class 'model_fit'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed model_fit object.

Examples

library(parsnip)
library(rpart)

# Create model and fit
lm_fit <- linear_reg() %>%
  set_engine("lm") %>%
  fit(mpg ~ ., data = mtcars)

out <- butcher(lm_fit, verbose = TRUE)

# Another parsnip model
rpart_fit <- decision_tree(mode = "regression") %>%
  set_engine("rpart") %>%
  fit(mpg ~ ., data = mtcars, minsplit = 5, cp = 0.1)

out <- butcher(rpart_fit, verbose = TRUE)

Axing an multnet.

Description

multnet objects are created from carrying out multinomial regression in the glmnet package.

Usage

## S3 method for class 'multnet'
axe_call(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed multnet object.

Examples

# Load libraries
library(parsnip)

# Load data
set.seed(1234)
predictrs <- matrix(rnorm(100*20), ncol = 20)
colnames(predictrs) <- paste0("a", seq_len(ncol(predictrs)))
response <- as.factor(sample(1:4, 100, replace = TRUE))

# Create model and fit
multnet_fit <- multinom_reg(penalty = 0.1) %>%
  set_engine("glmnet") %>%
  fit_xy(x = predictrs, y = response)

out <- butcher(multnet_fit, verbose = TRUE)

Axing a NaiveBayes.

Description

NaiveBayes objects are created from the klaR package, leveraged to fit a Naive Bayes Classifier.

Usage

## S3 method for class 'NaiveBayes'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'NaiveBayes'
axe_data(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed NaiveBayes object.

Examples

library(klaR)

fit_mod <- function() {
  boop <- runif(1e6)
  NaiveBayes(
    y ~ x,
    data = data.frame(y = as.factor(rep(letters[1:4], 1e4)), x = rnorm(4e4))
  )
}

mod_fit <- fit_mod()
mod_res <- butcher(mod_fit)

weigh(mod_fit)
weigh(mod_res)

Axing a nested_model_fit.

Description

nested_model_fit objects are created from the 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.

Usage

## S3 method for class 'nested_model_fit'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'nested_model_fit'
axe_ctrl(x, verbose = FALSE, ...)

## S3 method for class 'nested_model_fit'
axe_data(x, verbose = FALSE, ...)

## S3 method for class 'nested_model_fit'
axe_env(x, verbose = FALSE, ...)

## S3 method for class 'nested_model_fit'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed nested_model_fit object.

See Also

axe-model_fit

Examples

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)

Axing a nnet.

Description

nnet objects are created from the nnet package, leveraged to fit multilayer perceptron models.

Usage

## S3 method for class 'nnet'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'nnet'
axe_env(x, verbose = FALSE, ...)

## S3 method for class 'nnet'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed nnet object.

Examples

# Load libraries
library(parsnip)
library(nnet)

# Create and fit model
nnet_fit <- mlp("classification", hidden_units = 2) %>%
  set_engine("nnet") %>%
  fit(Species ~ ., data = iris)

out <- butcher(nnet_fit, verbose = TRUE)

# Another nnet object
targets <- class.ind(c(rep("setosa", 50),
                       rep("versicolor", 50),
                       rep("virginica", 50)))

fit <- nnet(iris[,1:4],
            targets,
            size = 2,
            rang = 0.1,
            decay = 5e-4,
            maxit = 20)

out <- butcher(fit, verbose = TRUE)

Axing mixOmics models

Description

mixo_pls (via pls()), mixo_spls (via spls()), and mixo_plsda (via plsda()) objects are created with the mixOmics package, leveraged to fit partial least squares models.

Usage

## S3 method for class 'mixo_pls'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'mixo_spls'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'mixo_pls'
axe_data(x, verbose = FALSE, ...)

## S3 method for class 'mixo_spls'
axe_data(x, verbose = FALSE, ...)

## S3 method for class 'mixo_pls'
axe_fitted(x, verbose = FALSE, ...)

## S3 method for class 'mixo_spls'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Details

The mixOmics package is not available on CRAN, but can be installed from the Bioconductor repository via remotes::install_bioc("mixOmics").

Value

Axed mixo_pls, mixo_spls, or mixo_plsda object.

Examples

library(butcher)
do.call(library, list(package = "mixOmics"))

# pls ------------------------------------------------------------------
fit_mod <- function() {
  boop <- runif(1e6)
  pls(matrix(rnorm(2e4), ncol = 2), rnorm(1e4), mode = "classic")
}

mod_fit <- fit_mod()
mod_res <- butcher(mod_fit)

weigh(mod_fit)
weigh(mod_res)

new_data <- matrix(1:2, ncol = 2)
colnames(new_data) <- c("X1", "X2")
predict(mod_fit, new_data)
predict(mod_res, new_data)

Axing an randomForest.

Description

randomForest objects are created from the randomForest package, which is used to train random forests based on Breiman's 2001 work. The package supports ensembles of classification and regression trees.

Usage

## S3 method for class 'randomForest'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'randomForest'
axe_ctrl(x, verbose = FALSE, ...)

## S3 method for class 'randomForest'
axe_env(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed randomForest object.

Examples

# Load libraries
library(parsnip)
library(rsample)
library(randomForest)
data(kyphosis, package = "rpart")

# Load data
set.seed(1234)
split <- initial_split(kyphosis, prop = 9/10)
spine_train <- training(split)

# Create model and fit
randomForest_fit <- rand_forest(mode = "classification",
                                mtry = 2,
                                trees = 2,
                                min_n = 3) %>%
  set_engine("randomForest") %>%
  fit_xy(x = spine_train[,2:4], y = spine_train$Kyphosis)

out <- butcher(randomForest_fit, verbose = TRUE)

# Another randomForest object
wrapped_rf <- function() {
  some_junk_in_environment <- runif(1e6)
  randomForest_fit <- randomForest(mpg ~ ., data = mtcars)
  return(randomForest_fit)
}

# Remove junk
cleaned_rf <- axe_env(wrapped_rf(), verbose = TRUE)

# Check size
lobstr::obj_size(cleaned_rf)

Axing an ranger.

Description

ranger objects are created from the ranger package, which is used as a means to quickly train random forests. The package supports ensembles of classification, regression, survival and probability prediction trees. Given the reliance of post processing functions on the model object, like importance_pvalues and treeInfo, on the first class listed, the butcher_ranger class is not appended.

Usage

## S3 method for class 'ranger'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'ranger'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed ranger object.

Examples

# Load libraries
library(parsnip)
library(rsample)
library(ranger)

# Load data
set.seed(1234)
split <- initial_split(iris, prop = 9/10)
iris_train <- training(split)

# Create model and fit
ranger_fit <- rand_forest(mode = "classification",
                          mtry = 2,
                          trees = 20,
                          min_n = 3) %>%
  set_engine("ranger") %>%
  fit(Species ~ ., data = iris_train)

out <- butcher(ranger_fit, verbose = TRUE)

# Another ranger object
wrapped_ranger <- function() {
  n <- 100
  p <- 400
  dat <- data.frame(y = factor(rbinom(n, 1, .5)), replicate(p, runif(n)))
  fit <- ranger(y ~ ., dat, importance = "impurity_corrected")
  return(fit)
}

cleaned_ranger <- axe_fitted(wrapped_ranger(), verbose = TRUE)

Axing an rda.

Description

rda objects are created from the klaR package, leveraged to carry out regularized discriminant analysis.

Usage

## S3 method for class 'rda'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'rda'
axe_env(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed rda object.

Examples

library(klaR)

fit_mod <- function() {
  boop <- runif(1e6)
  rda(
    y ~ x,
    data = data.frame(y = rep(letters[1:4], 1e4), x = rnorm(4e4)),
    gamma = 0.05,
    lambda = 0.2
  )
}

mod_fit <- fit_mod()
mod_res <- butcher(mod_fit)

weigh(mod_fit)
weigh(mod_res)

Axing a recipe object.

Description

recipe objects are created from the recipes package, which is leveraged for its set of data pre-processing tools. These recipes work by sequentially defining each pre-processing step. The implementation of each step, however, results its own class so we bundle all the axe methods related to recipe objects in general here. Note that the butchered class is only added to the recipe as a whole, and not to each pre-processing step.

Usage

## S3 method for class 'recipe'
axe_env(x, verbose = FALSE, ...)

## S3 method for class 'step'
axe_env(x, ...)

## S3 method for class 'step_arrange'
axe_env(x, ...)

## S3 method for class 'step_filter'
axe_env(x, ...)

## S3 method for class 'step_mutate'
axe_env(x, ...)

## S3 method for class 'step_slice'
axe_env(x, ...)

## S3 method for class 'step_impute_bag'
axe_env(x, ...)

## S3 method for class 'step_bagimpute'
axe_env(x, ...)

## S3 method for class 'step_impute_knn'
axe_env(x, ...)

## S3 method for class 'step_knnimpute'
axe_env(x, ...)

## S3 method for class 'step_geodist'
axe_env(x, ...)

## S3 method for class 'step_interact'
axe_env(x, ...)

## S3 method for class 'step_ratio'
axe_env(x, ...)

## S3 method for class 'quosure'
axe_env(x, ...)

## S3 method for class 'recipe'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed recipe object.

Examples

library(recipes)
data(biomass, package = "modeldata")

biomass_tr <- biomass[biomass$dataset == "Training",]
rec <- recipe(HHV ~ carbon + hydrogen + oxygen + nitrogen + sulfur,
              data = biomass_tr) %>%
  step_center(all_predictors()) %>%
  step_scale(all_predictors()) %>%
  step_spatialsign(all_predictors())

out <- butcher(rec, verbose = TRUE)

# Another recipe object
wrapped_recipes <- function() {
  some_junk_in_environment <- runif(1e6)
  return(
    recipe(mpg ~ cyl, data = mtcars) %>%
      step_center(all_predictors()) %>%
      step_scale(all_predictors()) %>%
      prep()
  )
}

# Remove junk in environment
cleaned1 <- axe_env(wrapped_recipes(), verbose = TRUE)
# Replace prepared training data with zero-row slice
cleaned2 <- axe_fitted(wrapped_recipes(), verbose = TRUE)

# Check size
lobstr::obj_size(cleaned1)
lobstr::obj_size(cleaned2)

Axing a rpart.

Description

rpart objects are created from the rpart package, which is used for recursive partitioning for classification, regression and survival trees.

Usage

## S3 method for class 'rpart'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'rpart'
axe_ctrl(x, verbose = FALSE, ...)

## S3 method for class 'rpart'
axe_data(x, verbose = FALSE, ...)

## S3 method for class 'rpart'
axe_env(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed rpart object.

Examples

# Load libraries
library(parsnip)
library(rsample)
library(rpart)

# Load data
set.seed(1234)
split <- initial_split(mtcars, prop = 9/10)
car_train <- training(split)

# Create model and fit
rpart_fit <- decision_tree(mode = "regression") %>%
  set_engine("rpart") %>%
  fit(mpg ~ ., data = car_train, minsplit = 5, cp = 0.1)

out <- butcher(rpart_fit, verbose = TRUE)

# Another rpart object
wrapped_rpart <- function() {
  some_junk_in_environment <- runif(1e6)
  fit <- rpart(Kyphosis ~ Age + Number + Start,
               data = kyphosis,
               x = TRUE, y = TRUE)
  return(fit)
}

# Remove junk
cleaned_rpart <- axe_env(wrapped_rpart(), verbose = TRUE)

# Check size
lobstr::obj_size(cleaned_rpart)

Axing a sclass object.

Description

sclass objects are byproducts of classbagg objects.

Usage

## S3 method for class 'sclass'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'sclass'
axe_env(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed sclass object.

Examples

# Load libraries
library(ipred)
library(rpart)
library(MASS)

# Load data
data("GlaucomaM", package = "TH.data")

classbagg_fit <- bagging(Class ~ ., data = GlaucomaM, coob = TRUE)

out <- butcher(classbagg_fit$mtrees[[1]], verbose = TRUE)

# Another classbagg object
wrapped_classbagg <- function() {
  some_junk_in_environment <- runif(1e6)
  fit <- bagging(Species ~ .,
                 data = iris,
                 nbagg = 10,
                 coob = TRUE)
  return(fit)
}

# Remove junk
cleaned_classbagg <- butcher(wrapped_classbagg(), verbose = TRUE)

# Check size
lobstr::obj_size(cleaned_classbagg)

Axing a spark object.

Description

spark objects are created from the sparklyr package, a R interface for Apache Spark. The axe methods available for spark objects are designed such that interoperability is maintained. In other words, for a multilingual machine learning team, butchered spark objects instantiated from sparklyr can still be serialized to disk, work in Python, be deployed on Scala, etc. It is also worth noting here that spark objects created from sparklyr have a lot of metadata attached to it, including but not limited to the formula, dataset, model, index labels, etc. The axe functions provided are for parsing down the model object both prior saving to disk, or loading from disk. Traditional R save functions are not available for these objects, so functionality is provided in sparklyr::ml_save. This function gives the user the option to keep either the pipeline_model or the pipeline, so both of these objects are retained from butchering, yet removal of one or the other might be conducive to freeing up memory on disk.

Usage

## S3 method for class 'ml_model'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'ml_model'
axe_ctrl(x, verbose = FALSE, ...)

## S3 method for class 'ml_model'
axe_data(x, verbose = FALSE, ...)

## S3 method for class 'ml_model'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed spark object.

Examples

library(sparklyr)

sc <- spark_connect(master = "local")

iris_tbls <- sdf_copy_to(sc, iris, overwrite = TRUE) %>%
  sdf_random_split(train = 2/3, validation = 2/3, seed = 2018)

train <- iris_tbls$train
spark_fit <- ml_logistic_regression(train, Species ~ .)

out <- butcher(spark_fit, verbose = TRUE)

spark_disconnect(sc)

Axing an survreg.

Description

survreg objects are created from the survival package. They are returned from the survreg function, representing fitted parametric survival models.

Usage

## S3 method for class 'survreg'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'survreg'
axe_data(x, verbose = FALSE, ...)

## S3 method for class 'survreg'
axe_env(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed survreg object.

Examples

# Load libraries
library(parsnip)
library(survival)

# Create model and fit
survreg_fit <- surv_reg(mode = "regression", dist = "weibull") %>%
  set_engine("survival") %>%
  fit(Surv(futime, fustat) ~ 1, data = ovarian)

out <- butcher(survreg_fit, verbose = TRUE)

# Another survreg object
wrapped_survreg <- function() {
  some_junk_in_environment <- runif(1e6)
  fit <- survreg(Surv(time, status) ~ ph.ecog + age + strata(sex),
                 data = lung)
  return(fit)
}

# Remove junk
cleaned_survreg <- butcher(wrapped_survreg(), verbose = TRUE)

# Check size
lobstr::obj_size(cleaned_survreg)

Axing an survreg.penal

Description

survreg.penal objects are created from the survival package. They are returned from the survreg function, representing fitted parametric survival models.

Usage

## S3 method for class 'survreg.penal'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'survreg.penal'
axe_data(x, verbose = FALSE, ...)

## S3 method for class 'survreg.penal'
axe_env(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed survreg object.

Examples

# Load libraries
library(parsnip)
library(survival)

# Create model and fit
survreg_fit <- surv_reg(mode = "regression", dist = "weibull") %>%
  set_engine("survival") %>%
  fit(Surv(time, status) ~ rx, data = rats)

out <- butcher(survreg_fit, verbose = TRUE)

# Another survreg.penal object
wrapped_survreg.penal <- function() {
  some_junk_in_environment <- runif(1e6)
  fit <- survreg(Surv(time, status) ~ rx,
                 data = rats, subset = (sex == "f"))
  return(fit)
}

# Remove junk
cleaned_sp <- axe_env(wrapped_survreg.penal(), verbose = TRUE)

# Check size
lobstr::obj_size(cleaned_sp)

Axing for terms inputs.

Description

Generics related to axing objects of the term class.

Usage

## S3 method for class 'terms'
axe_env(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed terms object.

Examples

# Using lm
wrapped_lm <- function() {
  some_junk_in_environment <- runif(1e6)
  fit <- lm(mpg ~ ., data = mtcars)
  return(fit)
}

# Remove junk
cleaned_lm <- axe_env(wrapped_lm(), verbose = TRUE)

# Check size
lobstr::obj_size(cleaned_lm)

# Compare environment in terms component
lobstr::obj_size(attr(wrapped_lm()$terms, ".Environment"))
lobstr::obj_size(attr(cleaned_lm$terms, ".Environment"))

# Using rpart
library(rpart)

wrapped_rpart <- function() {
  some_junk_in_environment <- runif(1e6)
  fit <- rpart(Kyphosis ~ Age + Number + Start,
               data = kyphosis,
               x = TRUE,
               y = TRUE)
  return(fit)
}

lobstr::obj_size(wrapped_rpart())
lobstr::obj_size(axe_env(wrapped_rpart()))

Axing a train object.

Description

train objects are created from the caret package.

Usage

## S3 method for class 'train'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'train'
axe_ctrl(x, verbose = FALSE, ...)

## S3 method for class 'train'
axe_data(x, verbose = FALSE, ...)

## S3 method for class 'train'
axe_env(x, verbose = FALSE, ...)

## S3 method for class 'train'
axe_fitted(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed train object.

Examples

# Load libraries
library(caret)

data(iris)
train_data <- iris[, 1:4]
train_classes <- iris[, 5]

train_fit <- train(train_data, train_classes,
                   method = "knn",
                   preProcess = c("center", "scale"),
                   tuneLength = 10,
                   trControl = trainControl(method = "cv"))

out <- butcher(train_fit, verbose = TRUE)

Axing a train.recipe object.

Description

train.recipe objects are slightly different from train objects created from the caret package in that it also includes instructions from a recipe for data pre-processing. Axing functions specific to train.recipe are thus included as additional steps are required to remove parts of train.recipe objects.

Usage

## S3 method for class 'train.recipe'
axe_call(x, ...)

## S3 method for class 'train.recipe'
axe_ctrl(x, ...)

## S3 method for class 'train.recipe'
axe_data(x, ...)

## S3 method for class 'train.recipe'
axe_env(x, ...)

## S3 method for class 'train.recipe'
axe_fitted(x, ...)

Arguments

x

A model object.
...

Any additional arguments related to axing.

Value

Axed train.recipe object.

Examples

library(recipes)
library(caret)
data(biomass, package = "modeldata")

data(biomass)
recipe <- biomass %>%
  recipe(HHV ~ carbon + hydrogen + oxygen + nitrogen + sulfur) %>%
  step_center(all_predictors()) %>%
  step_scale(all_predictors()) %>%
  step_spatialsign(all_predictors())

train.recipe_fit <- train(recipe, biomass,
                          method = "svmRadial",
                          metric = "RMSE")

out <- butcher(train.recipe_fit, verbose = TRUE)

Axing a xgb.Booster.

Description

xgb.Booster objects are created from the xgboost package, which provides efficient and scalable implementations of gradient boosted decision trees. Given the reliance of post processing functions on the model object, like xgb.Booster.complete, on the first class listed, the butcher_xgb.Booster class is not appended.

Usage

## S3 method for class 'xgb.Booster'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'xgb.Booster'
axe_env(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed xgb.Booster object.

Examples

library(xgboost)
library(parsnip)

data(agaricus.train)
bst <- xgboost(data = agaricus.train$data,
               label = agaricus.train$label,
               eta = 1,
               nthread = 2,
               nrounds = 2,
               eval_metric = "logloss",
               objective = "binary:logistic",
               verbose = 0)

out <- butcher(bst, verbose = TRUE)

# Another xgboost model
fit <- boost_tree(mode = "classification", trees = 20) %>%
  set_engine("xgboost", eval_metric = "mlogloss") %>%
  fit(Species ~ ., data = iris)

out <- butcher(fit, verbose = TRUE)

Axing a xrf.

Description

Axing a xrf.

Usage

## S3 method for class 'xrf'
axe_call(x, verbose = FALSE, ...)

## S3 method for class 'xrf'
axe_env(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed xrf object.

Examples

library(xrf)

xrf_big <- function() {
  boop <- runif(1e6)
  xrf(
    mpg ~ .,
    mtcars,
    xgb_control = list(nrounds = 2, max_depth = 2),
    family = 'gaussian'
  )
}

heavy_m <- xrf_big()

m <- butcher(heavy_m, verbose = TRUE)

weigh(heavy_m)
weigh(m)

Butcher an object.

Description

Reduce the size of a model object so that it takes up less memory on disk. Currently, the model object is stripped down to the point that only the minimal components necessary for the predict function to work remain. Future adjustments to this function will be needed to avoid removal of model fit components to ensure it works with other downstream functions.

Usage

butcher(x, verbose = FALSE, ...)

Arguments

x

A model object.
verbose

Print information each time an axe method is executed. Notes how much memory is released and what functions are disabled. Default is FALSE.
...

Any additional arguments related to axing.

Value

Axed model object with new butcher subclass assignment.

Locate part of an object.

Description

Locate where a specific component of a object might exist within the model object itself. This function is restricted in that only items that can be axed can be found.

Usage

locate(x, name = NULL)

Arguments

x

A model object.
name

A name associated with model component of interest. This defaults to NULL. Possible components include: env, call, data, ctrl, and fitted.

Value

Location of specific component in a model object.

Examples

lm_fit <- lm(mpg ~ ., data = mtcars)
locate(lm_fit, name = "env")
locate(lm_fit, name = "call")

New axe functions for a modeling object.

Description

new_model_butcher() will instantiate the following to help us develop new axe functions around removing parts of a new modeling object:

  • Add modeling package to Suggests

  • Generate and populate an axe file under R/

  • Generate and populate an test file under testthat/

Usage

new_model_butcher(model_class, package_name, open = interactive())

Arguments

model_class

A string that captures the class name of the new model object.
package_name

A string that captures the package name from which the new model is made.
open

Check if user is in interactive mode, and if so, opens the new files for editing.

Weigh the object.

Description

Evaluate the size of each element contained in a model object.

Usage

weigh(x, threshold = 0, units = "MB", ...)

Arguments

x

A model object.
threshold

The minimum threshold desired for model component size to display.
units

The units in which to display the size of each component within the model object of interest. Defaults to MB. Other options include KB and GB.
...

Any additional arguments for weighing.

Value

Tibble with weights of object components in decreasing magnitude.

Examples

simulate_x <- matrix(runif(1e+6), ncol = 2)
simulate_y <- runif(dim(simulate_x)[1])
lm_out <- lm(simulate_y ~ simulate_x)
weigh(lm_out)