Regression Stochastic Gradient Descent Learner

Stochastic Gradient Descent for learning various linear models. Calls RWeka::make_Weka_classifier() from RWeka.

Initial parameter values

• F:

  • Has only 3 out of 5 original loss functions: 2 = squared loss (regression), 3 = epsilon insensitive loss (regression) and 4 = Huber loss (regression) with 2 (squared loss) being the new default

  • Reason for change: this learner should only contain loss functions appropriate for regression tasks

Custom mlr3 parameters

• output_debug_info:

  • original id: output-debug-info

• do_not_check_capabilities:

  • original id: do-not-check-capabilities

• num_decimal_places:

  • original id: num-decimal-places

• batch_size:

  • original id: batch-size

• Reason for change: This learner contains changed ids of the following control arguments since their ids contain irregular pattern

Dictionary

This Learner can be instantiated via lrn():

lrn("regr.sgd")

Meta Information

• Task type: “regr”

• Predict Types: “response”

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

• Required Packages: mlr3, RWeka

Parameters

Id	Type	Default	Levels	Range
subset	untyped	-		-
na.action	untyped	-		-
F	character	2	2, 3, 4	-
L	numeric	0.01		\((-\infty, \infty)\)
R	numeric	1e-04		\((-\infty, \infty)\)
E	integer	500		\((-\infty, \infty)\)
C	numeric	0.001		\((-\infty, \infty)\)
N	logical	-	TRUE, FALSE	-
M	logical	-	TRUE, FALSE	-
S	integer	1		\((-\infty, \infty)\)
output_debug_info	logical	FALSE	TRUE, FALSE	-
do_not_check_capabilities	logical	FALSE	TRUE, FALSE	-
num_decimal_places	integer	2		\([1, \infty)\)
batch_size	integer	100		\([1, \infty)\)
options	untyped	NULL		-

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

damirpolat

Super classes

mlr3::Learner -> mlr3::LearnerRegr -> LearnerRegrSGD

Methods

Public methods

• LearnerRegrSGD$new()

• LearnerRegrSGD$clone()

Inherited methods

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

---

Method new()

Creates a new instance of this R6 class.

Usage

LearnerRegrSGD$new()

---

Method clone()

The objects of this class are cloneable with this method.

Usage

LearnerRegrSGD$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

Examples

# Define the Learner
learner = mlr3::lrn("regr.sgd")
print(learner)
#> <LearnerRegrSGD:regr.sgd>: Stochastic Gradient Descent
#> * Model: -
#> * Parameters: F=2
#> * Packages: mlr3, RWeka
#> * Predict Types:  [response]
#> * Feature Types: logical, integer, numeric, factor, ordered
#> * Properties: -

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

# 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)
#> Loss function: Squared loss (linear regression)
#> 
#> mpg = 
#> 
#>          0.6538 (normalized) am
#>  +      -4.0026 (normalized) carb
#>  +       3.433  (normalized) cyl
#>  +      -0.6412 (normalized) disp
#>  +       4.7144 (normalized) drat
#>  +       3.6504 (normalized) gear
#>  +      -4.626  (normalized) hp
#>  +       5.0495 (normalized) qsec
#>  +       0.6482 (normalized) vs
#>  +      -8.45   (normalized) wt
#>  +      18.0415


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

# Score the predictions
predictions$score()
#> regr.mse 
#>  9.09491