Type: Package
Title: Multi-Calibration Boosting
Version: 0.4.3
Description: Implements 'Multi-Calibration Boosting' (2018) https://proceedings.mlr.press/v80/hebert-johnson18a.html and 'Multi-Accuracy Boosting' (2019) <doi:10.48550/arXiv.1805.12317> for the multi-calibration of a machine learning model's prediction. 'MCBoost' updates predictions for sub-groups in an iterative fashion in order to mitigate biases like poor calibration or large accuracy differences across subgroups. Multi-Calibration works best in scenarios where the underlying data & labels are unbiased, but resulting models are. This is often the case, e.g. when an algorithm fits a majority population while ignoring or under-fitting minority populations.
License: LGPL (≥ 3)
URL: https://github.com/mlr-org/mcboost
BugReports: https://github.com/mlr-org/mcboost/issues
Encoding: UTF-8
Depends: R (≥ 3.1.0)
Imports: backports, checkmate (≥ 2.0.0), data.table (≥ 1.13.6), mlr3 (≥ 0.10), mlr3misc (≥ 0.8.0), mlr3pipelines (≥ 0.3.0), R6 (≥ 2.4.1), rmarkdown, rpart, glmnet
Suggests: curl, lgr, formattable, tidyverse, PracTools, mlr3learners, mlr3oml, neuralnet, paradox, knitr, ranger, xgboost, covr, testthat (≥ 3.1.0)
RoxygenNote: 7.3.1
VignetteBuilder: knitr
Collate: 'AuditorFitters.R' 'MCBoost.R' 'PipelineMCBoost.R' 'PipeOpLearnerPred.R' 'PipeOpMCBoost.R' 'Predictor.R' 'ProbRange.R' 'helpers.R' 'zzz.R'
NeedsCompilation: no
Packaged: 2024-04-10 19:32:07 UTC; sebi
Author: Florian Pfisterer ORCID iD [aut], Susanne Dandl ORCID iD [ctb], Christoph Kern ORCID iD [ctb], Carolin Becker [ctb], Bernd Bischl ORCID iD [ctb], Sebastian Fischer [ctb, cre]
Maintainer: Sebastian Fischer <sebf.fischer@gmail.com>
Repository: CRAN
Date/Publication: 2024-04-12 12:50:02 UTC

mcboost: Multi-Calibration Boosting

Description

Implements 'Multi-Calibration Boosting' (2018) https://proceedings.mlr.press/v80/hebert-johnson18a.html and 'Multi-Accuracy Boosting' (2019) doi:10.48550/arXiv.1805.12317 for the multi-calibration of a machine learning model's prediction. 'MCBoost' updates predictions for sub-groups in an iterative fashion in order to mitigate biases like poor calibration or large accuracy differences across subgroups. Multi-Calibration works best in scenarios where the underlying data & labels are unbiased, but resulting models are. This is often the case, e.g. when an algorithm fits a majority population while ignoring or under-fitting minority populations.

Author(s)

Maintainer: Sebastian Fischer sebf.fischer@gmail.com [contributor]

Authors:

Other contributors:

References

Kim et al., 2019: Multiaccuracy: Black-Box Post-Processing for Fairness in Classification. Hebert-Johnson et al., 2018: Multicalibration: Calibration for the (Computationally-Identifiable) Masses. Pfisterer F, Kern C, Dandl S, Sun M, Kim M, Bischl B (2021). “mcboost: Multi-Calibration Boosting for R.” Journal of Open Source Software, 6(64), 3453. doi:10.21105/joss.03453, https://joss.theoj.org/papers/10.21105/joss.03453.

See Also

Useful links:

AuditorFitter Abstract Base Class

Description

Defines an AuditorFitter abstract base class.

Value

list with items

Methods

Public methods

Method new()

Initialize a AuditorFitter. This is an abstract base class.

Usage
AuditorFitter$new()

Method fit_to_resid()

Fit to residuals.

Usage
AuditorFitter$fit_to_resid(data, resid, mask)
Arguments
data

data.table
Features.

resid

numeric
Residuals (of same length as data).

mask

integer
Mask applied to the data. Only used for SubgroupAuditorFitter.

Method fit()

Fit (mostly used internally, use fit_to_resid).

Usage
AuditorFitter$fit(data, resid, mask)
Arguments
data

data.table
Features.

resid

numeric
Residuals (of same length as data).

mask

integer
Mask applied to the data. Only used for SubgroupAuditorFitter.

Method clone()

The objects of this class are cloneable with this method.

Usage
AuditorFitter$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.

Cross-validated AuditorFitter from a Learner

Description

CVLearnerAuditorFitter returns the cross-validated predictions instead of the in-sample predictions.

Available data is cut into complementary subsets (folds). For each subset out-of-sample predictions are received by training a model on all other subsets and predicting afterwards on the left-out subset.

Value

AuditorFitter

list with items

Functions

Super class

mcboost::AuditorFitter -> CVLearnerAuditorFitter

Public fields

learner

CVLearnerPredictor
Learner used for fitting residuals.

Methods

Public methods

Inherited methods

Method new()

Define a CVAuditorFitter from a learner. Available instantiations:
CVTreeAuditorFitter (rpart) and CVRidgeAuditorFitter (glmnet). See mlr3pipelines::PipeOpLearnerCV for more information on cross-validated learners.

Usage
CVLearnerAuditorFitter$new(learner, folds = 3L)
Arguments
learner

mlr3::Learner
Regression Learner to use.

folds

integer
Number of folds to use for PipeOpLearnerCV. Defaults to 3.

Method fit()

Fit the cross-validated learner and compute correlation

Usage
CVLearnerAuditorFitter$fit(data, resid, mask)
Arguments
data

data.table
Features.

resid

numeric
Residuals (of same length as data).

mask

integer
Mask applied to the data. Only used for SubgroupAuditorFitter.

Method clone()

The objects of this class are cloneable with this method.

Usage
CVLearnerAuditorFitter$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.

Super classes

mcboost::AuditorFitter -> mcboost::CVLearnerAuditorFitter -> CVTreeAuditorFitter

Methods

Public methods

Inherited methods

Method new()

Define a cross-validated AuditorFitter from a rpart learner See mlr3pipelines::PipeOpLearnerCV for more information on cross-validated learners.

Usage
CVTreeAuditorFitter$new()

Method clone()

The objects of this class are cloneable with this method.

Usage
CVTreeAuditorFitter$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.

Super classes

mcboost::AuditorFitter -> mcboost::CVLearnerAuditorFitter -> CVRidgeAuditorFitter

Methods

Public methods

Inherited methods

Method new()

Define a cross-validated AuditorFitter from a glmnet learner. See mlr3pipelines::PipeOpLearnerCV for more information on cross-validated learners.

Usage
CVRidgeAuditorFitter$new()

Method clone()

The objects of this class are cloneable with this method.

Usage
CVRidgeAuditorFitter$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.

See Also

Other AuditorFitter: LearnerAuditorFitter, SubgroupAuditorFitter, SubpopAuditorFitter

Other AuditorFitter: LearnerAuditorFitter, SubgroupAuditorFitter, SubpopAuditorFitter

Other AuditorFitter: LearnerAuditorFitter, SubgroupAuditorFitter, SubpopAuditorFitter

Create an AuditorFitter from a Learner

Description

Instantiates an AuditorFitter that trains a mlr3::Learner on the data.

Value

AuditorFitter

list with items

Functions

Super class

mcboost::AuditorFitter -> LearnerAuditorFitter

Public fields

learner

LearnerPredictor
Learner used for fitting residuals.

Methods

Public methods

Inherited methods

Method new()

Define an AuditorFitter from a Learner. Available instantiations:
TreeAuditorFitter (rpart) and RidgeAuditorFitter (glmnet).

Usage
LearnerAuditorFitter$new(learner)
Arguments
learner

mlr3::Learner
Regression learner to use.

Method fit()

Fit the learner and compute correlation

Usage
LearnerAuditorFitter$fit(data, resid, mask)
Arguments
data

data.table
Features.

resid

numeric
Residuals (of same length as data).

mask

integer
Mask applied to the data. Only used for SubgroupAuditorFitter.

Method clone()

The objects of this class are cloneable with this method.

Usage
LearnerAuditorFitter$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.

Super classes

mcboost::AuditorFitter -> mcboost::LearnerAuditorFitter -> TreeAuditorFitter

Methods

Public methods

Inherited methods

Method new()

Define a AuditorFitter from a rpart learner.

Usage
TreeAuditorFitter$new()

Method clone()

The objects of this class are cloneable with this method.

Usage
TreeAuditorFitter$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.

Super classes

mcboost::AuditorFitter -> mcboost::LearnerAuditorFitter -> RidgeAuditorFitter

Methods

Public methods

Inherited methods

Method new()

Define a AuditorFitter from a glmnet learner.

Usage
RidgeAuditorFitter$new()

Method clone()

The objects of this class are cloneable with this method.

Usage
RidgeAuditorFitter$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.

See Also

Other AuditorFitter: CVLearnerAuditorFitter, SubgroupAuditorFitter, SubpopAuditorFitter

Other AuditorFitter: CVLearnerAuditorFitter, SubgroupAuditorFitter, SubpopAuditorFitter

Other AuditorFitter: CVLearnerAuditorFitter, SubgroupAuditorFitter, SubpopAuditorFitter

Multi-Calibration Boosting

Description

Implements Multi-Calibration Boosting by Hebert-Johnson et al. (2018) and Multi-Accuracy Boosting by Kim et al. (2019) for the multi-calibration of a machine learning model's prediction. Multi-Calibration works best in scenarios where the underlying data & labels are unbiased but a bias is introduced within the algorithm's fitting procedure. This is often the case, e.g. when an algorithm fits a majority population while ignoring or under-fitting minority populations.
Expects initial models that fit binary outcomes or continuous outcomes with predictions that are in (or scaled to) the 0-1 range. The method defaults to ⁠Multi-Accuracy Boosting⁠ as described in Kim et al. (2019). In order to obtain behaviour as described in Hebert-Johnson et al. (2018) set multiplicative=FALSE and num_buckets to 10.

For additional details, please refer to the relevant publications:

Public fields

max_iter

integer
The maximum number of iterations of the multi-calibration/multi-accuracy method.

alpha

numeric
Accuracy parameter that determines the stopping condition.

eta

numeric
Parameter for multiplicative weight update (step size).

num_buckets

integer
The number of buckets to split into in addition to using the whole sample.

bucket_strategy

character
Currently only supports "simple", even split along probabilities. Only relevant for num_buckets > 1.

rebucket

logical
Should buckets be re-calculated at each iteration?

eval_fulldata

logical
Should auditor be evaluated on the full data?

partition

logical
True/False flag for whether to split up predictions by their "partition" (e.g., predictions less than 0.5 and predictions greater than 0.5).

multiplicative

logical
Specifies the strategy for updating the weights (multiplicative weight vs additive).

iter_sampling

character
Specifies the strategy to sample the validation data for each iteration.

auditor_fitter

AuditorFitter
Specifies the type of model used to fit the residuals.

predictor

function
Initial predictor function.

iter_models

list
Cumulative list of fitted models.

iter_partitions

list
Cumulative list of data partitions for models.

iter_corr

list
Auditor correlation in each iteration.

auditor_effects

list
Auditor effect in each iteration.

bucket_strategies

character
Possible bucket_strategies.

weight_degree

integer
Weighting degree for low-degree multi-calibration.

Methods

Public methods

Method new()

Initialize a multi-calibration instance.

Usage
MCBoost$new(
  max_iter = 5,
  alpha = 1e-04,
  eta = 1,
  num_buckets = 2,
  partition = ifelse(num_buckets > 1, TRUE, FALSE),
  bucket_strategy = "simple",
  rebucket = FALSE,
  eval_fulldata = FALSE,
  multiplicative = TRUE,
  auditor_fitter = NULL,
  subpops = NULL,
  default_model_class = ConstantPredictor,
  init_predictor = NULL,
  iter_sampling = "none",
  weight_degree = 1L
)
Arguments
max_iter

integer
The maximum number of iterations of the multi-calibration/multi-accuracy method. Default 5L.

alpha

numeric
Accuracy parameter that determines the stopping condition. Default 1e-4.

eta

numeric
Parameter for multiplicative weight update (step size). Default 1.0.

num_buckets

integer
The number of buckets to split into in addition to using the whole sample. Default 2L.

partition

logical
True/False flag for whether to split up predictions by their "partition" (e.g., predictions less than 0.5 and predictions greater than 0.5). Defaults to TRUE (multi-accuracy boosting).

bucket_strategy

character
Currently only supports "simple", even split along probabilities. Only taken into account for num_buckets > 1.

rebucket

logical
Should buckets be re-done at each iteration? Default FALSE.

eval_fulldata

logical
Should the auditor be evaluated on the full data or on the respective bucket for determining the stopping criterion? Default FALSE, auditor is only evaluated on the bucket. This setting keeps the implementation closer to the Algorithm proposed in the corresponding multi-accuracy paper (Kim et al., 2019) where auditor effects are computed across the full sample (i.e. eval_fulldata = TRUE).

multiplicative

logical
Specifies the strategy for updating the weights (multiplicative weight vs additive). Defaults to TRUE (multi-accuracy boosting). Set to FALSE for multi-calibration.

auditor_fitter

AuditorFitter|character|mlr3::Learner
Specifies the type of model used to fit the residuals. The default is RidgeAuditorFitter. Can be a character, the name of a AuditorFitter, a mlr3::Learner that is then auto-converted into a LearnerAuditorFitter or a custom AuditorFitter.

subpops

list
Specifies a collection of characteristic attributes and the values they take to define subpopulations e.g. list(age = c('20-29','30-39','40+'), nJobs = c(0,1,2,'3+'), ,..).

default_model_class

Predictor
The class of the model that should be used as the init predictor model if init_predictor is not specified. Defaults to ConstantPredictor which predicts a constant value.

init_predictor

function|mlr3::Learner
The initial predictor function to use (i.e., if the user has a pretrained model). If a mlr3 Learner is passed, it will be autoconverted using mlr3_init_predictor. This requires the mlr3::Learner to be trained.

iter_sampling

character
How to sample the validation data for each iteration? Can be bootstrap, split or none.
"split" splits the data into max_iter parts and validates on each sample in each iteration.
"bootstrap" uses a new bootstrap sample in each iteration.
"none" uses the same dataset in each iteration.

weight_degree

character
Weighting degree for low-degree multi-calibration. Initialized to 1, which applies constant weighting with 1.

Method multicalibrate()

Run multi-calibration.

Usage
MCBoost$multicalibrate(data, labels, predictor_args = NULL, audit = FALSE, ...)
Arguments
data

data.table
Features.

labels

numeric
One-hot encoded labels (of same length as data).

predictor_args

any
Arguments passed on to init_predictor. Defaults to NULL.

audit

logical
Perform auditing? Initialized to TRUE.

...

any
Params passed on to other methods.

Returns

NULL

Method predict_probs()

Predict a dataset with multi-calibrated predictions

Usage
MCBoost$predict_probs(x, t = Inf, predictor_args = NULL, audit = FALSE, ...)
Arguments
x

data.table
Prediction data.

t

integer
Number of multi-calibration steps to predict. Default: Inf (all).

predictor_args

any
Arguments passed on to init_predictor. Defaults to NULL.

audit

logical
Should audit weights be stored? Default FALSE.

...

any
Params passed on to the residual prediction model's predict method.

Returns

numeric
Numeric vector of multi-calibrated predictions.

Method auditor_effect()

Compute the auditor effect for each instance which are the cumulative absolute predictions of the auditor. It indicates "how much" each observation was affected by multi-calibration on average across iterations.

Usage
MCBoost$auditor_effect(
  x,
  aggregate = TRUE,
  t = Inf,
  predictor_args = NULL,
  ...
)
Arguments
x

data.table
Prediction data.

aggregate

logical
Should the auditor effect be aggregated across iterations? Defaults to TRUE.

t

integer
Number of multi-calibration steps to predict. Defaults to Inf (all).

predictor_args

any
Arguments passed on to init_predictor. Defaults to NULL.

...

any
Params passed on to the residual prediction model's predict method.

Returns

numeric
Numeric vector of auditor effects for each row in x.

Method print()

Prints information about multi-calibration.

Usage
MCBoost$print(...)
Arguments
...

any
Not used.

Method clone()

The objects of this class are cloneable with this method.

Usage
MCBoost$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.

Examples

# See vignette for more examples.
# Instantiate the object
## Not run: 
mc = MCBoost$new()
# Run multi-calibration on training dataset.
mc$multicalibrate(iris[1:100, 1:4], factor(sample(c("A", "B"), 100, TRUE)))
# Predict on test set
mc$predict_probs(iris[101:150, 1:4])
# Get auditor effect
mc$auditor_effect(iris[101:150, 1:4])

## End(Not run)

Static AuditorFitter based on Subgroups

Description

Used to assess multi-calibration based on a list of binary subgroup_masks passed during initialization.

Value

AuditorFitter

list with items

Super class

mcboost::AuditorFitter -> SubgroupAuditorFitter

Public fields

subgroup_masks

list
List of subgroup masks. Initialize a SubgroupAuditorFitter

Methods

Public methods

Inherited methods

Method new()

Initializes a SubgroupAuditorFitter that assesses multi-calibration within each group defined by the ‘subpops’.

Usage
SubgroupAuditorFitter$new(subgroup_masks)
Arguments
subgroup_masks

list
List of subgroup masks. Subgroup masks are list(s) of integer masks, each with the same length as data to be fitted on. They allow defining subgroups of the data.

Method fit()

Fit the learner and compute correlation

Usage
SubgroupAuditorFitter$fit(data, resid, mask)
Arguments
data

data.table
Features.

resid

numeric
Residuals (of same length as data).

mask

integer
Mask applied to the data. Only used for SubgroupAuditorFitter.

Method clone()

The objects of this class are cloneable with this method.

Usage
SubgroupAuditorFitter$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.

See Also

Other AuditorFitter: CVLearnerAuditorFitter, LearnerAuditorFitter, SubpopAuditorFitter

Examples

 library("data.table")
 data = data.table(
   "AGE_0_10" =  c(1, 1, 0, 0, 0),
   "AGE_11_20" = c(0, 0, 1, 0, 0),
   "AGE_21_31" = c(0, 0, 0, 1, 1),
   "X1" = runif(5),
   "X2" = runif(5)
 )
 label = c(1,0,0,1,1)
 masks = list(
   "M1" = c(1L, 0L, 1L, 1L, 0L),
   "M2" = c(1L, 0L, 0L, 0L, 1L)
 )
 sg = SubgroupAuditorFitter$new(masks)

Static AuditorFitter based on Subpopulations

Description

Used to assess multi-calibration based on a list of binary valued columns: subpops passed during initialization.

Value

AuditorFitter

list with items

Super class

mcboost::AuditorFitter -> SubpopAuditorFitter

Public fields

subpops

list
List of subpopulation indicators. Initialize a SubpopAuditorFitter

Methods

Public methods

Inherited methods

Method new()

Initializes a SubpopAuditorFitter that assesses multi-calibration within each group defined by the ⁠subpops'. Names in ⁠subpops' must correspond to columns in the data.

Usage
SubpopAuditorFitter$new(subpops)
Arguments
subpops

list
Specifies a collection of characteristic attributes and the values they take to define subpopulations e.g. list(age = c('20-29','30-39','40+'), nJobs = c(0,1,2,'3+'), ,..).

Method fit()

Fit the learner and compute correlation

Usage
SubpopAuditorFitter$fit(data, resid, mask)
Arguments
data

data.table
Features.

resid

numeric
Residuals (of same length as data).

mask

integer
Mask applied to the data. Only used for SubgroupAuditorFitter.

Method clone()

The objects of this class are cloneable with this method.

Usage
SubpopAuditorFitter$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.

See Also

Other AuditorFitter: CVLearnerAuditorFitter, LearnerAuditorFitter, SubgroupAuditorFitter

Examples

  library("data.table")
  data = data.table(
    "AGE_NA" = c(0, 0, 0, 0, 0),
    "AGE_0_10" =  c(1, 1, 0, 0, 0),
    "AGE_11_20" = c(0, 0, 1, 0, 0),
    "AGE_21_31" = c(0, 0, 0, 1, 1),
    "X1" = runif(5),
    "X2" = runif(5)
  )
  label = c(1,0,0,1,1)
  pops = list("AGE_NA", "AGE_0_10", "AGE_11_20", "AGE_21_31", function(x) {x[["X1" > 0.5]]})
  sf = SubpopAuditorFitter$new(subpops = pops)
  sf$fit(data, label - 0.5)

Create an initial predictor function from a trained mlr3 learner

Description

Create an initial predictor function from a trained mlr3 learner

Usage

mlr3_init_predictor(learner)

Arguments

learner

mlr3::Learner A trained learner used for initialization.

Value

function

Examples

 ## Not run: 
 library("mlr3")
 l = lrn("classif.featureless")$train(tsk("sonar"))
 mlr3_init_predictor(l)
 
## End(Not run)

Multi-Calibrate a Learner's Prediction

Description

mlr3pipelines::PipeOp that trains a Learner and passes its predictions forward during training and prediction.

Post-process a learner prediction using multi-calibration. For more details, please refer to https://arxiv.org/pdf/1805.12317.pdf (Kim et al. 2018) or the help for MCBoost. If no init_predictor is provided, the preceding learner's predictions corresponding to the prediction slot are used as an initial predictor for MCBoost.

Format

R6Class inheriting from mlr3pipelines::PipeOp.

R6Class inheriting from mlr3pipelines::PipeOp.

Construction

PipeOpLearnerPred$new(learner, id = NULL, param_vals = list())

* `learner` :: [`Learner`][mlr3::Learner] \cr
  [`Learner`][mlr3::Learner] to  prediction, or a string identifying a
  [`Learner`][mlr3::Learner] in the [`mlr3::mlr_learners`] [`Dictionary`][mlr3misc::Dictionary].
* `id` :: `character(1)`
  Identifier of the resulting object, internally defaulting to the `id` of the [`Learner`][mlr3::Learner] being wrapped.
* `param_vals` :: named `list`\cr
  List of hyperparameter settings, overwriting the hyperparameter settings that would otherwise be set during construction. Default `list()`.


[mlr3::Learner]: R:mlr3::Learner
[mlr3::Learner]: R:mlr3::Learner
[mlr3::Learner]: R:mlr3::Learner
[`mlr3::mlr_learners`]: R:%60mlr3::mlr_learners%60
[mlr3misc::Dictionary]: R:mlr3misc::Dictionary
[mlr3::Learner]: R:mlr3::Learner

PipeOpMCBoost$new(id = "mcboost", param_vals = list())

Input and Output Channels

PipeOpLearnerPred has one input channel named "input", taking a Task specific to the Learner type given to learner during construction; both during training and prediction.

PipeOpLearnerPred has one output channel named "output", producing a Task specific to the Learner type given to learner during construction; both during training and prediction.

During training, the input and output are "data" and "prediction", two TaskClassif. A PredictionClassif is required as input and returned as output during prediction.

State

The ⁠$state⁠ is a MCBoost Object as obtained from MCBoost$new().

Parameters

The ⁠$state⁠ is set to the ⁠$state⁠ slot of the Learner object, together with the ⁠$state⁠ elements inherited from mlr3pipelines::PipeOpTaskPreproc. It is a named list with the inherited members, as well as:

Fields

Fields inherited from PipeOp, as well as:

Only fields inherited from mlr3pipelines::PipeOp.

Methods

Methods inherited from mlr3pipelines::PipeOpTaskPreproc/mlr3pipelines::PipeOp.

Only methods inherited from mlr3pipelines::PipeOp.

Super classes

mlr3pipelines::PipeOp -> mlr3pipelines::PipeOpTaskPreproc -> PipeOpLearnerPred

Active bindings

learner

The wrapped learner.

learner_model

The wrapped learner's model(s).

Methods

Public methods

Inherited methods

Method new()

Initialize a Learner Predictor PipeOp. Can be used to wrap trained or untrainted mlr3 learners.

Usage
PipeOpLearnerPred$new(learner, id = NULL, param_vals = list())
Arguments
learner

Learner
The learner that should be wrapped.

id

character
The PipeOp's id. Defaults to "mcboost".

param_vals

list
List of hyperparameters for the PipeOp.

Method clone()

The objects of this class are cloneable with this method.

Usage
PipeOpLearnerPred$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.

Super class

mlr3pipelines::PipeOp -> PipeOpMCBoost

Active bindings

predict_type

Predict type of the PipeOp.

Methods

Public methods

Inherited methods

Method new()

Initialize a Multi-Calibration PipeOp.

Usage
PipeOpMCBoost$new(id = "mcboost", param_vals = list())
Arguments
id

character
The PipeOp's id. Defaults to "mcboost".

param_vals

list
List of hyperparameters for the PipeOp.

Method clone()

The objects of this class are cloneable with this method.

Usage
PipeOpMCBoost$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.

See Also

https://mlr3book.mlr-org.com/list-pipeops.html

https://mlr3book.mlr-org.com/list-pipeops.html

Examples

## Not run: 
gr = gunion(list(
  "data" = po("nop"),
  "prediction" = po("learner_cv", lrn("classif.rpart"))
)) %>>%
  PipeOpMCBoost$new()
tsk = tsk("sonar")
tid = sample(1:208, 108)
gr$train(tsk$clone()$filter(tid))
gr$predict(tsk$clone()$filter(setdiff(1:208, tid)))

## End(Not run)

One-hot encode a factor variable

Description

One-hot encode a factor variable

Usage

one_hot(labels)

Arguments

labels

factor
Factor to encode.

Value

integer
Integer vector of encoded labels.

Examples

 ## Not run: 
 one_hot(factor(c("a", "b", "a")))
 
## End(Not run)

Multi-calibration pipeline

Description

Wraps MCBoost in a Pipeline to be used with mlr3pipelines. For now this assumes training on the same dataset that is later used for multi-calibration.

Usage

ppl_mcboost(learner = lrn("classif.featureless"), param_vals = list())

Arguments

learner

(mlr3)mlr3::Learner
Initial learner. Internally wrapped into a PipeOpLearnerCV with resampling.method = "insample" as a default. All parameters can be adjusted through the resulting Graph's param_set. Defaults to lrn("classif.featureless"). Note: An initial predictor can also be supplied via the init_predictor parameter.

param_vals

list
List of parameter values passed on to MCBoost$new.

Value

(mlr3pipelines) Graph

Examples

  ## Not run: 
  library("mlr3pipelines")
  gr = ppl_mcboost()
  
## End(Not run)