Relative Feature Importance

Implementation of RFI using modular sampling approach

References

König, Gunnar, Molnar, Christoph, Bischl, Bernd, Grosse-Wentrup, Moritz (2021). “Relative Feature Importance.” In 2020 25th International Conference on Pattern Recognition (ICPR), 9318–9325. doi:10.1109/ICPR48806.2021.9413090 .

Super classes

xplainfi::FeatureImportanceMethod -> xplainfi::PerturbationImportance -> RFI

Methods

Public methods

• RFI$new()

• RFI$compute()

• RFI$clone()

Inherited methods

• xplainfi::FeatureImportanceMethod$combine()
• xplainfi::FeatureImportanceMethod$print()
• xplainfi::FeatureImportanceMethod$reset()

---

Method new()

Creates a new instance of the RFI class

Usage

RFI$new(
  task,
  learner,
  measure,
  resampling = NULL,
  features = NULL,
  conditioning_set = NULL,
  relation = "difference",
  iters_perm = 1L,
  sampler = NULL
)

Arguments

task, learner, measure, resampling, features

Passed to PerturbationImportance

conditioning_set

(character()) Set of features to condition on. Can be overridden in $compute(). Default (character(0)) is equivalent to PFI. In CFI, this would be set to all features except tat of interest.

relation

(character(1)) How to relate perturbed scores to originals. Can be overridden in $compute().

iters_perm

(integer(1)) Number of permutation iterations. Can be overridden in $compute().

sampler

(ConditionalSampler) Optional custom sampler. Defaults to ARFSampler

---

Method compute()

Compute RFI scores

Usage

RFI$compute(
  relation = NULL,
  conditioning_set = NULL,
  iters_perm = NULL,
  store_backends = TRUE
)

Arguments

relation

(character(1)) How to relate perturbed scores to originals. If NULL, uses stored value.

conditioning_set

(character()) Set of features to condition on. If NULL, uses the stored parameter value.

iters_perm

(integer(1)) Number of permutation iterations. If NULL, uses stored value.

store_backends

(logical(1)) Whether to store backends

---

Method clone()

The objects of this class are cloneable with this method.

Usage

RFI$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

Examples

library(mlr3)
task = tgen("friedman1")$generate(n = 200)
rfi = RFI$new(
  task = task,
  learner = lrn("regr.ranger", num.trees = 50),
  measure = msr("regr.mse"),
  conditioning_set = c("important1")
)
#> ℹ No <ConditionalSampler> provided, using <ARFSampler> with default settings.
#> ℹ No <Resampling> provided, using holdout resampling with default ratio.
rfi$compute()
#> Key: <feature>
#>          feature  importance
#>           <char>       <num>
#>  1:   important1  0.00000000
#>  2:   important2  4.56934410
#>  3:   important3  1.03210376
#>  4:   important4  9.62226686
#>  5:   important5  2.45917494
#>  6: unimportant1  0.84629886
#>  7: unimportant2  0.16659014
#>  8: unimportant3 -0.23757522
#>  9: unimportant4  0.06427863
#> 10: unimportant5 -0.55808231