Boosted Generalized Linear Classification Learner

Fit a generalized linear classification model using a boosting algorithm. Calls mboost::glmboost() from mboost.

Dictionary

This Learner can be instantiated via lrn():

lrn("classif.glmboost")

Meta Information

• Task type: “classif”

• Predict Types: “response”, “prob”

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

• Required Packages: mlr3, mlr3extralearners, mboost

Parameters

Id	Type	Default	Levels	Range
offset	numeric	NULL		\((-\infty, \infty)\)
family	character	Binomial	Binomial, AdaExp, AUC, custom	-
custom.family	untyped	-		-
link	character	logit	logit, probit	-
type	character	adaboost	glm, adaboost	-
center	logical	TRUE	TRUE, FALSE	-
mstop	integer	100		\((-\infty, \infty)\)
nu	numeric	0.1		\((-\infty, \infty)\)
risk	character	inbag	inbag, oobag, none	-
oobweights	untyped	NULL		-
trace	logical	FALSE	TRUE, FALSE	-
stopintern	untyped	FALSE		-
na.action	untyped	stats::na.omit		-
contrasts.arg	untyped	-		-

References

Bühlmann, Peter, Yu, Bin (2003). “Boosting with the L 2 loss: regression and classification.” Journal of the American Statistical Association, 98(462), 324–339.

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

be-marc

Super classes

mlr3::Learner -> mlr3::LearnerClassif -> LearnerClassifGLMBoost

Methods

Public methods

• LearnerClassifGLMBoost$new()

• LearnerClassifGLMBoost$clone()

Inherited methods

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

---

Method new()

Create a LearnerClassifGLMBoost object.

Usage

LearnerClassifGLMBoost$new()

---

Method clone()

The objects of this class are cloneable with this method.

Usage

LearnerClassifGLMBoost$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

Examples

# Define the Learner
learner = mlr3::lrn("classif.glmboost")
print(learner)
#> <LearnerClassifGLMBoost:classif.glmboost>: Boosted Generalized Linear Model
#> * Model: -
#> * Parameters: list()
#> * Packages: mlr3, mlr3extralearners, mboost
#> * Predict Types:  [response], prob
#> * Feature Types: integer, numeric, factor, ordered
#> * Properties: twoclass, weights

# 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)
#> 
#> 	 Generalized Linear Models Fitted via Gradient Boosting
#> 
#> Call:
#> glmboost.formula(formula = f, data = data, family = new("boost_family_glm",     fW = function (f)     {        f <- pmin(abs(f), 36) * sign(f)        p <- exp(f)/(exp(f) + exp(-f))        4 * p * (1 - p)    }, ngradient = function (y, f, w = 1)     {        exp2yf <- exp(-2 * y * f)        -(-2 * y * exp2yf)/(log(2) * (1 + exp2yf))    }, risk = function (y, f, w = 1)     sum(w * loss(y, f), na.rm = TRUE), offset = function (y,         w)     {        p <- weighted.mean(y > 0, w)        1/2 * log(p/(1 - p))    }, check_y = function (y)     {        if (!is.factor(y))             stop("response is not a factor but ", sQuote("family = Binomial()"))        if (nlevels(y) != 2)             stop("response is not a factor at two levels but ",                 sQuote("family = Binomial()"))        return(c(-1, 1)[as.integer(y)])    }, weights = function (w)     {        switch(weights, any = TRUE, none = isTRUE(all.equal(unique(w),             1)), zeroone = isTRUE(all.equal(unique(w + abs(w -             1)), 1)), case = isTRUE(all.equal(unique(w - floor(w)),             0)))    }, nuisance = function ()     return(NA), response = function (f)     {        f <- pmin(abs(f), 36) * sign(f)        p <- exp(f)/(exp(f) + exp(-f))        return(p)    }, rclass = function (f)     (f > 0) + 1, name = "Negative Binomial Likelihood (logit link)",     charloss = c("{ \n", "    f <- pmin(abs(f), 36) * sign(f) \n",     "    p <- exp(f)/(exp(f) + exp(-f)) \n", "    y <- (y + 1)/2 \n",     "    -y * log(p) - (1 - y) * log(1 - p) \n", "} \n")), control = ctrl)
#> 
#> 
#> 	 Negative Binomial Likelihood (logit link) 
#> 
#> Loss function: { 
#>      f <- pmin(abs(f), 36) * sign(f) 
#>      p <- exp(f)/(exp(f) + exp(-f)) 
#>      y <- (y + 1)/2 
#>      -y * log(p) - (1 - y) * log(1 - p) 
#>  } 
#>  
#> 
#> Number of boosting iterations: mstop = 100 
#> Step size:  0.1 
#> Offset:  -0.1375516 
#> 
#> Coefficients: 
#> 
#> NOTE: Coefficients from a Binomial model are half the size of coefficients
#>  from a model fitted via glm(... , family = 'binomial').
#> See Warning section in ?coef.mboost
#> (Intercept)          V1         V11         V12         V15         V16 
#>  1.44850329 -1.61973244 -1.98614906 -0.51676485  0.27994821  0.24371062 
#>         V20         V21         V22         V23         V31         V36 
#> -0.12385778 -0.05242839 -0.57733875 -0.16141872  0.21417983  0.97756917 
#>          V4         V40         V43         V45         V47         V48 
#> -1.60584671  0.29314894 -0.84063438 -1.29701959 -0.39839278 -2.94077397 
#>         V49         V52         V55         V56 
#> -0.82349669 -6.25062747  3.12775491  1.87245764 
#> attr(,"offset")
#> [1] -0.1375516
#> 


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

# Score the predictions
predictions$score()
#> classif.ce 
#>  0.2028986