Package 'broomstick'

Tidying methods for a randomForest model

Description

These methods tidy the variable importance of a random forest model summary, augment the original data with information on the fitted values/classifications and error, and construct a one-row glance of the model's statistics.

Usage

## S3 method for class 'randomForest'
augment(x, data = NULL, ...)

## S3 method for class 'randomForest'
glance(x, ...)

## S3 method for class 'randomForest'
tidy(x, ...)

Arguments

x	randomForest object
data	Model data for use by augment.randomForest().
...	Additional arguments (ignored)

Value

augment.randomForest returns the original data with additional columns:

.oob_times	The number of trees for which the given case was "out of bag". See randomForest::randomForest() for more details.
.fitted	The fitted value or class.

augment returns additional columns for classification and usupervised trees:

.votes	For each case, the voting results, with one column per class.
.local_var_imp	The casewise variable importance, stored as data frames in a nested list-column, with one row per variable in the model. Only present if the model was created with importance = TRUE

glance.randomForest returns a data.frame with the following columns for regression trees:

mse	The average mean squared error across all trees.
rsq	The average pesudo-R-squared across all trees. See randomForest::randomForest() for more information.

For classification trees: one row per class, with the following columns:

precision
recall
accuracy
f_measure

All tidying methods return a data.frame without rownames. The structure depends on the method chosen.

tidy.randomForest returns one row for each model term, with the following columns:

term	The term in the randomForest model
MeanDecreaseAccuracy	A measure of variable importance. See randomForest::randomForest() for more information. Only present if the model was created with importance = TRUE
MeanDecreaseGini	A measure of variable importance. See randomForest::randomForest() for more information.
MeanDecreaseAccuracy_sd	Standard deviation of MeanDecreaseAccuracy. See randomForest::randomForest() for more information. Only present if the model was created with importance = TRUE
classwise_importance	Classwise variable importance for each term, stored as data frames in a nested list-column, with one row per class. Only present if the model was created with importance = TRUE

---

Augment your model object

Description

Augment your model object

Usage

## S3 method for class 'rpart'
augment(x, data = NULL, newdata = NULL, ...)

Arguments

x	rpart model
data	data.frame from the model
newdata	new data to use for predictions, residuals, etc.
...	extra arguments to pass

Value

augment.rpart returns the original data with additional columns:

• .fitted: The fitted value or class.

• .resid: only given when the same data as was used for the model is provided.

Examples

library(rpart)
rpart_fit <- rpart(Sepal.Width ~ ., iris)
augment(rpart_fit)

---

Convert Decision Tree Analysis Objects into Tidy Data Frames

Description

Convert decision tree analysis objects from R into tidy data frames, so that they can more easily be combined, reshaped and otherwise processed with tools like dplyr, tidyr and ggplot2. The package provides three S3 generics: tidy, which summarizes a model's statistical findings such as coefficients of a regression; augment, which adds columns to the original data such as predictions, residuals and cluster assignments; and glance, which provides a one-row summary of model-level statistics.

---

tidy up the model summary of gbm

Description

tidy returns a tibble of variable importance for the rpart pacakge

Usage

## S3 method for class 'gbm'
tidy(x, n_trees = x$n.trees, scale = FALSE, sort = TRUE, normalise = TRUE, ...)

Arguments

x	A gbm model
n_trees	integer. (optional) Number of trees to use for computing relative importance. Default is the number of trees in x$n.trees. If not provided, a guess is made using the heuristic: If a test set was used in fitting, the number of trees resulting in lowest test set error will be used; else, if cross-validation was performed, the number of trees resulting in lowest cross-validation error will be used; otherwise, all trees will be used.
scale	(optional) Should importance be scaled? Default is FALSE
sort	(optional) Should results be sorted? Default is TRUE
normalise	(optional) Should results be normalised to sum to 100? Default is TRUE
...	extra functions or arguments

Value

A tibble containing the importance score for each variable

Examples

# retrieve a tibble of the variable importance from an gbm model

library(gbm)
library(MASS)
fit_gbm <- gbm(calories ~., data = UScereal)

tidy(fit_gbm)

---

tidy up the model summary of rpart

Description

tidy returns a tibble of variable importance for the rpart pacakge

Usage

## S3 method for class 'rpart'
tidy(x, ...)

Arguments

x	An rpart model
...	extra functions or arguments

Value

A tibble containing the importance score for each variable

Examples

# retrieve a tibble of the variable importance from an rpart model

library(rpart)
fit_rpart <- rpart(Kyphosis ~ Age + Number + Start, data = kyphosis)

tidy(fit_rpart)

---