Classification MARS (Multivariate Adaptive Regression Splines) Learner

This is an alternative implementation of MARS (Multivariate Adaptive Regression Splines). The classification problem is solved by 0-1 encoding of the two-class targets and setting the decision threshold to p = 0.5 during the prediction phase. MARS is trademarked and thus not used as the name. The name "earth" stands for "Enhanced Adaptive Regression Through Hinges".

Details

Methods for variance estimations are not yet implemented.

Dictionary

This Learner can be instantiated via lrn():

lrn("classif.earth")

Meta Information

• Task type: “classif”

• Predict Types: “response”, “prob”

• Feature Types: “integer”, “numeric”, “factor”

• Required Packages: mlr3, mlr3extralearners, earth

Parameters

Id	Type	Default	Levels	Range
wp	untyped	NULL		-
offset	untyped	NULL		-
keepxy	logical	FALSE	TRUE, FALSE	-
trace	character	0	0, .3, .5, 1, 2, 3, 4, 5	-
degree	integer	1		\([1, \infty)\)
penalty	numeric	2		\([-1, \infty)\)
nk	untyped	NULL		-
thresh	numeric	0.001		\((-\infty, \infty)\)
minspan	numeric	0		\([0, \infty)\)
endspan	numeric	0		\([0, \infty)\)
newvar.penalty	numeric	0		\([0, \infty)\)
fast.k	integer	20		\([0, \infty)\)
fast.beta	integer	1		\([0, 1]\)
linpreds	untyped	FALSE		-
allowed	untyped	-		-
pmethod	character	backward	backward, none, exhaustive, forward, seqrep, cv	-
nprune	integer	-		\([0, \infty)\)
nfold	integer	0		\([0, \infty)\)
ncross	integer	1		\([0, \infty)\)
stratify	logical	TRUE	TRUE, FALSE	-
varmod.method	character	none	none, const, lm, rlm, earth, gam, power, power0, x.lm, x.rlm, ...	-
varmod.exponent	numeric	1		\((-\infty, \infty)\)
varmod.conv	numeric	1		\([0, 1]\)
varmod.clamp	numeric	0.1		\((-\infty, \infty)\)
varmod.minspan	numeric	-3		\((-\infty, \infty)\)
Scale.y	logical	FALSE	TRUE, FALSE	-
Adjust.endspan	numeric	2		\((-\infty, \infty)\)
Auto.linpreds	logical	TRUE	TRUE, FALSE	-
Force.weights	logical	FALSE	TRUE, FALSE	-
Use.beta.cache	logical	TRUE	TRUE, FALSE	-
Force.xtx.prune	logical	FALSE	TRUE, FALSE	-
Get.leverages	logical	TRUE	TRUE, FALSE	-
Exhaustive.tol	numeric	1e-10		\((-\infty, \infty)\)

References

Milborrow, Stephen, Hastie, T, Tibshirani, R (2014). “Earth: multivariate adaptive regression spline models.” R package version, 3(7).

Friedman, H J (1991). “Multivariate adaptive regression splines.” The annals of statistics, 19(1), 1–67.

See also

• Dictionary of Learners: mlr3::mlr_learners.

• as.data.table(mlr_learners) for a table of available Learners in the running session (depending on the loaded packages).

• Chapter in the mlr3book: https://mlr3book.mlr-org.com/basics.html#learners

• mlr3learners for a selection of recommended learners.

• mlr3cluster for unsupervised clustering learners.

• mlr3pipelines to combine learners with pre- and postprocessing steps.

• mlr3tuning for tuning of hyperparameters, mlr3tuningspaces for established default tuning spaces.

Author

pkopper

Super classes

mlr3::Learner -> mlr3::LearnerClassif -> LearnerClassifEarth

Methods

Public methods

• LearnerClassifEarth$new()

• LearnerClassifEarth$clone()

Inherited methods

• mlr3::Learner$base_learner()
• mlr3::Learner$configure()
• mlr3::Learner$encapsulate()
• mlr3::Learner$format()
• mlr3::Learner$help()
• mlr3::Learner$predict()
• mlr3::Learner$predict_newdata()
• mlr3::Learner$print()
• mlr3::Learner$reset()
• mlr3::Learner$selected_features()
• mlr3::Learner$train()

---

Method new()

Creates a new instance of this R6 class.

Usage

LearnerClassifEarth$new()

---

Method clone()

The objects of this class are cloneable with this method.

Usage

LearnerClassifEarth$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

Examples

# Define the Learner
learner = mlr3::lrn("classif.earth")
print(learner)
#> 
#> ── <LearnerClassifEarth> (classif.earth): Multivariate Adaptive Splines ────────
#> • Model: -
#> • Parameters: list()
#> • Packages: mlr3, mlr3extralearners, and earth
#> • Predict Types: [response] and prob
#> • Feature Types: integer, numeric, and factor
#> • Encapsulation: none (fallback: -)
#> • Properties: twoclass and weights
#> • Other settings: use_weights = 'use'

# Define a Task
task = mlr3::tsk("sonar")

# Create train and test set
ids = mlr3::partition(task)

# Train the learner on the training ids
learner$train(task, row_ids = ids$train)

print(learner$model)
#> GLM (family binomial, link logit):
#>  nulldev  df       dev  df   devratio     AIC iters converged
#>  192.112 138   59.1966 128      0.692    81.2     7         1
#> 
#> Earth selected 11 of 22 terms, and 9 of 60 predictors
#> Termination condition: RSq changed by less than 0.001 at 22 terms
#> Importance: V11, V36, V43, V23, V25, V46, V60, V27, V21-unused, V47, ...
#> Number of terms at each degree of interaction: 1 10 (additive model)
#> Earth GCV 0.1269697    RSS 12.7189    GRSq 0.4972949    RSq 0.6324477


# Make predictions for the test rows
predictions = learner$predict(task, row_ids = ids$test)

# Score the predictions
predictions$score()
#> classif.ce 
#>  0.2318841