Package 'fmeffects'

Description

Computes forward marginal effects (FME) for arbitrary supervised machine learning models. You provide a model and data, and 'fmeffects' gives you feature effects.

Author(s)

Maintainer: Holger Löwe [email protected]

Authors:

• Christian Scholbeck [email protected]

Other contributors:

• Christian Heumann [email protected] [reviewer]

• Bernd Bischl [email protected] [reviewer]

• Giuseppe Casalicchio [email protected] [reviewer]

See Also

Useful links:

• https://holgstr.github.io/fmeffects/

• https://github.com/holgstr/fmeffects

• Report bugs at https://github.com/holgstr/fmeffects/issues

Description

This is a wrapper function for AverageMarginalEffects$new(...)$compute(). It computes Average Marginal Effects (AME) based on Forward Marginal Effects (FME) for a model. The AME is a simple mean FME and computed w.r.t. a feature variable and a model.

Usage

ame(model, data, features = NULL, ep.method = "none")

Arguments

Value

An AverageMarginalEffects object, with a field results containing a list of summary statistics, including

• Feature: The name of the feature.

• step.size: The step.size w.r.t. the specified feature.

• AME: The Average Marginal Effect for a step of length step.size w.r.t. the specified feature.

• SD: The standard deviation of FMEs for the specified feature and step.size.

• 0.25: The 0.25-quantile of FMEs for the specified feature and step.size.

• 0.75: The 0.75-quantile of FMEs for the specified feature and step.size.

• n: The number of observations included for the computation of the AME. This can vary for the following reasons: For categorical features, FMEs are only computed for observations where the original category is not the step.size category. For numerical features, FMEs are only computed for observations that are not extrapolation points (if ep.method is set to "envelope").

References

Scholbeck, C.A., Casalicchio, G., Molnar, C. et al. Marginal effects for non-linear prediction functions. Data Min Knowl Disc (2024). https://doi.org/10.1007/s10618-023-00993-x

Examples

# Train a model:

library(mlr3verse)
library(ranger)
data(bikes, package = "fmeffects")
set.seed(123)
task = as_task_regr(x = bikes, id = "bikes", target = "count")
forest = lrn("regr.ranger")$train(task)

# Compute AMEs for all features:
## Not run: 
overview = ame(model = forest, data = bikes)
summary(overview)

# Compute AMEs for a subset of features with non-default step.sizes:
overview = ame(model = forest,
               data = bikes,
               features = list(humidity = 0.1, weather = c("clear", "rain")))
summary(overview)

# Extract results:
overview$results

## End(Not run)

Description

The AME is a simple mean FME and computed w.r.t. a feature variable and a model.

Public fields

Methods

Public methods

• AverageMarginalEffects$new()

• AverageMarginalEffects$compute()

• AverageMarginalEffects$clone()

Method new()

Create a new AME object.

Usage

AverageMarginalEffects$new(model, data, features = NULL, ep.method = "none")

Arguments

Returns

A new AME object.

Examples

# Train a model:

library(mlr3verse)
library(ranger)
set.seed(123)
data(bikes, package = "fmeffects")
task = as_task_regr(x = bikes, id = "bikes", target = "count")
forest = lrn("regr.ranger")$train(task)

# Compute AMEs for all features:
\dontrun{
overview = AverageMarginalEffects$new(
  model = forest,
  data = bikes)$compute()
summary(overview)

# Compute AMEs for a subset of features with non-default step.sizes:
overview = AverageMarginalEffects$new(model = forest,
                                      data = bikes,
                                      features = list(humidity = 0.1,
                                                   weather = c("clear", "rain")))$compute()
summary(overview)
}

Method compute()

Computes results, i.e., AMEs including the SD of FMEs, for an AME object.

Usage

AverageMarginalEffects$compute()

Returns

An AME object with results.

Examples

# Compute results:
\dontrun{
overview$compute()
}

Method clone()

The objects of this class are cloneable with this method.

Usage

AverageMarginalEffects$clone(deep = FALSE)

Arguments

Examples

## ------------------------------------------------
## Method `AverageMarginalEffects$new`
## ------------------------------------------------

# Train a model:

library(mlr3verse)
library(ranger)
set.seed(123)
data(bikes, package = "fmeffects")
task = as_task_regr(x = bikes, id = "bikes", target = "count")
forest = lrn("regr.ranger")$train(task)

# Compute AMEs for all features:
## Not run: 
overview = AverageMarginalEffects$new(
  model = forest,
  data = bikes)$compute()
summary(overview)

# Compute AMEs for a subset of features with non-default step.sizes:
overview = AverageMarginalEffects$new(model = forest,
                                      data = bikes,
                                      features = list(humidity = 0.1,
                                                   weather = c("clear", "rain")))$compute()
summary(overview)

## End(Not run)

## ------------------------------------------------
## Method `AverageMarginalEffects$compute`
## ------------------------------------------------

# Compute results:
## Not run: 
overview$compute()

## End(Not run)

Description

This data set contains information on daily bike sharing usage in Washington, D.C. for the years 2011-2012. The target variable is count, the total number of bikes lent out to users at a specific day.

Usage

data(bikes)

Format

An object of class data.frame with 731 rows and 10 columns.

Details

This data frame contains the following columns:

season

Season of the year

year

Year; 0=2011, 1=2012

holiday

If a day is a public holiday (y/n)

weekday

Day of the week

workingday

If a day is aworking day (y/n)

weather

Weather situation

temp

Temperature in degrees celsius

humidity

Humidity (relative)

windspeed

Windspeed in miles per hour

count

Total number of bikes lent out to users

Source

The original data can be found on the UCI database (ID = 275).

References

Fanaee-T, Hadi, and Gama, Joao, "Event labeling combining ensemble detectors and background knowledge", Progress in Artificial Intelligence (2013): pp. 1-15, Springer Berlin Heidelberg, doi:10.1007/s13748-013-0040-3.

Description

This is a wrapper function that creates the correct subclass of Partitioning. It computes feature subspaces for semi-global interpretations of FMEs via recursive partitioning (RP).

Usage

came(
  effects,
  number.partitions = NULL,
  max.sd = Inf,
  rp.method = "ctree",
  tree.control = NULL
)

Arguments

Value

Partitioning Object with identified feature subspaces.

References

Scholbeck, C.A., Casalicchio, G., Molnar, C. et al. Marginal effects for non-linear prediction functions. Data Min Knowl Disc (2024). https://doi.org/10.1007/s10618-023-00993-x

Examples

# Train a model and compute FMEs:

library(mlr3verse)
library(ranger)
data(bikes, package = "fmeffects")
task = as_task_regr(x = bikes, id = "bikes", target = "count")
forest = lrn("regr.ranger")$train(task)
effects = fme(model = forest, data = bikes, features = list("temp" = 1), ep.method = "envelope")

# Find a partitioning with exactly 3 subspaces:
subspaces = came(effects, number.partitions = 3)

# Find a partitioning with a maximum standard deviation of 20, use `rpart`:
library(rpart)
subspaces = came(effects, max.sd = 200, rp.method = "rpart")

# Analyze results:
summary(subspaces)
plot(subspaces)

# Extract results:
subspaces$results
subspaces$tree

Description

This is a wrapper function for FME$new(...)$compute(). It computes forward marginal effects (FMEs) for a specified change in feature values.

Usage

fme(
  model,
  data,
  features,
  ep.method = "none",
  compute.nlm = FALSE,
  nlm.intervals = 1
)

Arguments

Details

If one or more numeric features are passed to the features argument, FMEs are computed as

$FME_{x, h_{S}} = f(x + h_{S}, x_{-S}) - f(x)$

where $h_{S}$ is the step size vector and $x_{-S}$ the other features. If one or more categorical features are passed to features,

$FME_{x, c_{J}} = f(c_{J}, x_{-J}) - f(x)$

where $c_{J}$ is the set of selected reference categories in features and $x_{-J}$ the other features.

Value

ForwardsMarginalEffect object with the following fields:

• ame average marginal effect (AME).

• anlm average non-linearity measure (NLM).

• extrapolation.ids observations that have been identified as extrapolation points and not included in the analysis.

• data.step, a data.table of the feature matrix after the step has been applied.

• results, a data.table of the individual FMEs (and NLMs, if applicable) for all observations that are not extrapolation points.

References

Scholbeck, C.A., Casalicchio, G., Molnar, C. et al. Marginal effects for non-linear prediction functions. Data Min Knowl Disc (2024). https://doi.org/10.1007/s10618-023-00993-x

Examples

# Train a model:

library(mlr3verse)
library(ranger)
data(bikes, package = "fmeffects")
forest = lrn("regr.ranger")$train(as_task_regr(x = bikes, target = "count"))

# Compute FMEs for a numerical feature:
effects = fme(model = forest, data = bikes, features = list("temp" = 1), ep.method = "envelope")

# Analyze results:
summary(effects)
plot(effects)

# Extract results:
effects$results

# Compute the AME for a categorial feature:
fme(model = forest, data = bikes, features = list("weather" = "rain"))$ame

Description

The FME is a forward difference in prediction due to a specified change in feature values.

Public fields

Methods

Public methods

• ForwardMarginalEffect$new()

• ForwardMarginalEffect$compute()

• ForwardMarginalEffect$plot()

• ForwardMarginalEffect$clone()

Method new()

Create a new ForwardMarginalEffect object.

Usage

ForwardMarginalEffect$new(
  predictor,
  features,
  ep.method = "none",
  compute.nlm = FALSE,
  nlm.intervals = 1
)

Arguments

Returns

A new ForwardMarginalEffect object.

Examples

# Train a model:

library(mlr3verse)
library(ranger)
data(bikes, package = "fmeffects")
forest = lrn("regr.ranger")$train(as_task_regr(x = bikes, target = "count"))

# Create an `ForwardMarginalEffect` object:
effects = ForwardMarginalEffect$new(makePredictor(forest, bikes),
                  features = list("temp" = 1, "humidity" = 0.01),
                  ep.method = "envelope")

Method compute()

Computes results, i.e., FME (and NLMs) for non-extrapolation points, for a ForwardMarginalEffect object.

Usage

ForwardMarginalEffect$compute()

Returns

A ForwardMarginalEffect object with results.

Examples

# Compute results:
effects$compute()

Method plot()

Plots results, i.e., FME (and NLMs) for non-extrapolation points, for an FME object.

Usage

ForwardMarginalEffect$plot(with.nlm = FALSE, bins = 40, binwidth = NULL)

Arguments

Examples

# Compute results:
effects$plot()

Method clone()

The objects of this class are cloneable with this method.

Usage

ForwardMarginalEffect$clone(deep = FALSE)

Arguments

Examples

## ------------------------------------------------
## Method `ForwardMarginalEffect$new`
## ------------------------------------------------


# Train a model:

library(mlr3verse)
library(ranger)
data(bikes, package = "fmeffects")
forest = lrn("regr.ranger")$train(as_task_regr(x = bikes, target = "count"))

# Create an `ForwardMarginalEffect` object:
effects = ForwardMarginalEffect$new(makePredictor(forest, bikes),
                  features = list("temp" = 1, "humidity" = 0.01),
                  ep.method = "envelope")

## ------------------------------------------------
## Method `ForwardMarginalEffect$compute`
## ------------------------------------------------

# Compute results:
effects$compute()

## ------------------------------------------------
## Method `ForwardMarginalEffect$plot`
## ------------------------------------------------

# Compute results:
effects$plot()

User-friendly function to create a Predictor.

Description

A wrapper function that creates the correct subclass of Predictor by automatically from model. Can be passed to the constructor of FME.

Usage

makePredictor(model, data)

Arguments

Examples

# Train a model:

library(mlr3verse)
data(bikes, package = "fmeffects")
task = as_task_regr(x = bikes, id = "bikes", target = "count")
forest = lrn("regr.ranger")$train(task)

# Create the predictor:
predictor = makePredictor(forest, bikes)

# This instantiated an object of the correct subclass of `Predictor`:
class(predictor)

Description

This is the abstract superclass for partitioning objects like PartitioningCtree and PartitioningRpart. A Partitioning contains information about feature subspaces with conditional average marginal effects (cAME) computed for ForwardMarginalEffect objects.

Public fields

Methods

Public methods

• Partitioning$new()

• Partitioning$compute()

• Partitioning$plot()

• Partitioning$clone()

Method new()

Create a Partitioning object

Usage

Partitioning$new(...)

Arguments

Method compute()

Computes the partitioning, i.e., feature subspaces with more homogeneous FMEs, for a ForwardMarginalEffect object.

Usage

Partitioning$compute()

Returns

An Partitioning object with results.

Examples

# Compute results for an arbitrary partitioning:
# subspaces$compute()

Method plot()

Plots results, i.e., a decision tree and summary statistics of the feature subspaces, for an Partitioning object after ⁠$compute()⁠ has been run.

Usage

Partitioning$plot()

Examples

# Plot an arbitrary partitioning:
# subspaces$plot()

Method clone()

The objects of this class are cloneable with this method.

Usage

Partitioning$clone(deep = FALSE)

Arguments

Examples

## ------------------------------------------------
## Method `Partitioning$compute`
## ------------------------------------------------

# Compute results for an arbitrary partitioning:
# subspaces$compute()

## ------------------------------------------------
## Method `Partitioning$plot`
## ------------------------------------------------

# Plot an arbitrary partitioning:
# subspaces$plot()

Description

This task specializes Partitioning for the ctree algorithm for recursive partitioning.

It is recommended to use came() for construction of Partitioning objects.

Super class

fmeffects::Partitioning -> PartitioningCtree

Methods

Public methods

• PartitioningCtree$new()

• PartitioningCtree$clone()

Method new()

Create a new PartitioningCtree object.

Usage

PartitioningCtree$new(object, method, value, tree.control = NULL)

Arguments

Method clone()

The objects of this class are cloneable with this method.

Usage

PartitioningCtree$clone(deep = FALSE)

Arguments

Description

This task specializes Partitioning for the rpart algorithm for recursive partitioning.

It is recommended to use came() for construction of Partitioning objects.

Super class

fmeffects::Partitioning -> PartitioningRpart

Methods

Public methods

• PartitioningRpart$new()

• PartitioningRpart$clone()

Method new()

Create a new PartitioningRpart object.

Usage

PartitioningRpart$new(object, method, value, tree.control = NULL)

Arguments

Method clone()

The objects of this class are cloneable with this method.

Usage

PartitioningRpart$clone(deep = FALSE)

Arguments

Description

Plots an ForwardMarginalEffect object.

Usage

## S3 method for class 'ForwardMarginalEffect'
plot(x, ...)

Arguments

Description

Plots an FME Partitioning.

Usage

## S3 method for class 'Partitioning'
plot(x, ...)

Arguments

Description

This is the abstract superclass for predictor objects like PredictorMLR3 and PredictorCaret. A Predictor contains information about an ML model's prediction function and training data.

Public fields

Methods

Public methods

• Predictor$new()

• Predictor$clone()

Method new()

Create a Predictor object

Usage

Predictor$new(...)

Arguments

Method clone()

The objects of this class are cloneable with this method.

Usage

Predictor$clone(deep = FALSE)

Arguments

Description

This task specializes Predictor for caret regression models. The model is assumed to be a c("train", "train.formula").

It is recommended to use makePredictor() for construction of Predictor objects.

Super class

fmeffects::Predictor -> PredictorCaret

Methods

Public methods

• PredictorCaret$new()

• PredictorCaret$predict()

• PredictorCaret$clone()

Method new()

Create a new PredictorCaret object.

Usage

PredictorCaret$new(model, data)

Arguments

Method predict()

Predicts on an observation "newdata".

Usage

PredictorCaret$predict(newdata)

Arguments

Method clone()

The objects of this class are cloneable with this method.

Usage

PredictorCaret$clone(deep = FALSE)

Arguments

Description

This task specializes Predictor for lm and lm-type models. The model is assumed to be a lm.

It is recommended to use makePredictor() for construction of Predictor objects.

Super class

fmeffects::Predictor -> PredictorLM

Methods

Public methods

• PredictorLM$new()

• PredictorLM$predict()

• PredictorLM$clone()

Method new()

Create a new PredictorCaret object.

Usage

PredictorLM$new(model, data)

Arguments

Method predict()

Predicts on an observation "newdata".

Usage

PredictorLM$predict(newdata)

Arguments

Method clone()

The objects of this class are cloneable with this method.

Usage

PredictorLM$clone(deep = FALSE)

Arguments

Description

This task specializes Predictor for mlr3 models. The model is assumed to be a LearnerRegr or LearnerClassif.

It is recommended to use makePredictor() for construction of Predictor objects.

Super class

fmeffects::Predictor -> PredictorMLR3

Methods

Public methods

• PredictorMLR3$new()

• PredictorMLR3$predict()

• PredictorMLR3$clone()

Method new()

Create a new PredictorMLR3 object.

Usage

PredictorMLR3$new(model, data)

Arguments

Method predict()

Predicts on an observation "newdata".

Usage

PredictorMLR3$predict(newdata)

Arguments

Method clone()

The objects of this class are cloneable with this method.

Usage

PredictorMLR3$clone(deep = FALSE)

Arguments

Description

This task specializes Predictor for parsnip models. The model is assumed to be a model_fit object.

It is recommended to use makePredictor() for construction of Predictor objects.

Super class

fmeffects::Predictor -> PredictorParsnip

Methods

Public methods

• PredictorParsnip$new()

• PredictorParsnip$predict()

• PredictorParsnip$clone()

Method new()

Create a new PredictorParsnip object.

Usage

PredictorParsnip$new(model, data)

Arguments

Method predict()

Predicts on an observation "newdata".

Usage

PredictorParsnip$predict(newdata)

Arguments

Method clone()

The objects of this class are cloneable with this method.

Usage

PredictorParsnip$clone(deep = FALSE)

Arguments

Description

Prints an ForwardMarginalEffect object.

Usage

## S3 method for class 'ForwardMarginalEffect'
print(x, ...)

Arguments

Description

Prints an FME Partitioning.

Usage

## S3 method for class 'Partitioning'
print(x, ...)

Arguments

Description

Prints summary of an AverageMarginalEffects object.

Usage

## S3 method for class 'AverageMarginalEffects'
summary(object, ...)

Arguments

Description

Prints summary of an ForwardMarginalEffect object.

Usage

## S3 method for class 'ForwardMarginalEffect'
summary(object, ...)

Arguments

Description

Prints summary of an FME Partitioning.

Usage

## S3 method for class 'Partitioning'
summary(object, ...)

Arguments