| Title: | RF Variable Importance for Arbitrary Measures |
| Version: | 0.4 |
| Description: | Computes the random forest variable importance (VIMP) for the conditional inference random forest (cforest) of the 'party' package. Includes a function (varImp) that computes the VIMP for arbitrary measures from the 'measures' package. For calculating the VIMP regarding the measures accuracy and AUC two extra functions exist (varImpACC and varImpAUC). |
| License: | GPL-3 |
| Encoding: | UTF-8 |
| Depends: | R (≥ 3.0.0), measures, party, stats |
| LazyData: | true |
| Date: | 2020-05-15 |
| RoxygenNote: | 7.0.2 |
| Suggests: | testthat, ranger |
| NeedsCompilation: | no |
| Packaged: | 2020-05-23 08:01:54 UTC; philipp |
| Author: | Philipp Probst [aut, cre], Silke Janitza [ctb] |
| Maintainer: | Philipp Probst <philipp_probst@gmx.de> |
| Repository: | CRAN |
| Date/Publication: | 2020-05-23 09:00:02 UTC |
varImp
Description
Computes the variable importance for arbitrary measures from the 'measures' package.
Usage
varImp(
object,
mincriterion = 0,
conditional = FALSE,
threshold = 0.2,
nperm = 1,
OOB = TRUE,
pre1.0_0 = conditional,
measure = "multiclass.Brier",
...
)
Arguments
object |
An object as returned by cforest. |
mincriterion |
The value of the test statistic or 1 - p-value that must be exceeded in order to include a split in the computation of the importance. The default mincriterion = 0 guarantees that all splits are included. |
conditional |
a logical determining whether unconditional or conditional computation of the importance is performed. |
threshold |
The threshold value for (1 - p-value) of the association between the variable of interest and a covariate, which must be exceeded inorder to include the covariate in the conditioning scheme for the variable of interest (only relevant if conditional = TRUE). A threshold value of zero includes all covariates. |
nperm |
The number of permutations performed. |
OOB |
A logical determining whether the importance is computed from the out-of-bag sample or the learning sample (not suggested). |
pre1.0_0 |
Prior to party version 1.0-0, the actual data values were permuted according to the original permutation importance suggested by Breiman (2001). Now the assignments to child nodes of splits in the variable of interest are permuted as described by Hapfelmeier et al. (2012), which allows for missing values in the explanatory variables and is more efficient wrt memory consumption and computing time. This method does not apply to conditional variable importances. |
measure |
The name of the measure of the 'measures' package that should be used for the variable importance calculation. |
... |
Further arguments (like positive or negativ class) that are needed by the measure. |
Details
Many measures have not been tested for the usefulness of random forests variable importance. Use at your own risk.
Value
Vector with computed permutation importance for each variable.
Examples
# multiclass case
data(iris)
iris.cf = cforest(Species ~ ., data = iris, control = cforest_unbiased(mtry = 2, ntree = 50))
set.seed(123)
vimp = varImp(object = iris.cf, measure = "multiclass.Brier")
vimp
varImpACC
Description
Computes the variable importance regarding the accuracy (ACC).
Usage
varImpACC(
object,
mincriterion = 0,
conditional = FALSE,
threshold = 0.2,
nperm = 1,
OOB = TRUE,
pre1.0_0 = conditional
)
Arguments
object |
An object as returned by cforest. |
mincriterion |
The value of the test statistic or 1 - p-value that must be exceeded in order to include a split in the computation of the importance. The default mincriterion = 0 guarantees that all splits are included. |
conditional |
The value of the test statistic or 1 - p-value that must be exceeded in order to include a split in the computation of the importance. The default mincriterion = 0 guarantees that all splits are included. |
threshold |
The threshold value for (1 - p-value) of the association between the variable of interest and a covariate, which must be exceeded inorder to include the covariate in the conditioning scheme for the variable of interest (only relevant if conditional = TRUE). A threshold value of zero includes all covariates. |
nperm |
The number of permutations performed. |
OOB |
A logical determining whether the importance is computed from the out-of-bag sample or the learning sample (not suggested). |
pre1.0_0 |
Prior to party version 1.0-0, the actual data values were permuted according to the original permutation importance suggested by Breiman (2001). Now the assignments to child nodes of splits in the variable of interest are permuted as described by Hapfelmeier et al. (2012), which allows for missing values in the explanatory variables and is more efficient wrt memory consumption and computing time. This method does not apply to conditional variable importances. |
Value
Vector with computed permutation importance for each variable
Examples
data(iris)
iris2 = iris
iris2$Species = factor(iris$Species == "versicolor")
iris.cf = cforest(Species ~ ., data = iris2,control = cforest_unbiased(mtry = 2, ntree = 50))
set.seed(123)
a = varImpACC(object = iris.cf)
varImpAUC
Description
Computes the variable importance regarding the AUC. Bindings are not taken into account in the AUC definition as they did not provide as good results as the version without bindings in the paper of Janitza et. al (2013) (see References section).
Usage
varImpAUC(
object,
mincriterion = 0,
conditional = FALSE,
threshold = 0.2,
nperm = 1,
OOB = TRUE,
pre1.0_0 = conditional
)
Arguments
object |
An object as returned by cforest. |
mincriterion |
The value of the test statistic or 1 - p-value that must be exceeded in order to include a split in the computation of the importance. The default mincriterion = 0 guarantees that all splits are included. |
conditional |
The value of the test statistic or 1 - p-value that must be exceeded in order to include a split in the computation of the importance. The default mincriterion = 0 guarantees that all splits are included. |
threshold |
The threshold value for (1 - p-value) of the association between the variable of interest and a covariate, which must be exceeded inorder to include the covariate in the conditioning scheme for the variable of interest (only relevant if conditional = TRUE). A threshold value of zero includes all covariates. |
nperm |
The number of permutations performed. |
OOB |
A logical determining whether the importance is computed from the out-of-bag sample or the learning sample (not suggested). |
pre1.0_0 |
Prior to party version 1.0-0, the actual data values were permuted according to the original permutation importance suggested by Breiman (2001). Now the assignments to child nodes of splits in the variable of interest are permuted as described by Hapfelmeier et al. (2012), which allows for missing values in the explanatory variables and is more efficient wrt memory consumption and computing time. This method does not apply to conditional variable importances. |
Details
For using the original AUC definition and multiclass AUC you can use the varImp function and specify the particular measure.
Value
Vector with computed permutation importance for each variable
References
https://bmcbioinformatics.biomedcentral.com/articles/10.1186/1471-2105-14-119
Examples
# multiclass case
data(iris)
iris2 = iris
iris2$Species = factor(iris$Species == "versicolor")
iris.cf = cforest(Species ~ ., data = iris2,control = cforest_unbiased(mtry = 2, ntree = 50))
set.seed(123)
varImpAUC(object = iris.cf)
varImpRanger
Description
Computes the variable importance for ranger models and for arbitrary measures from the 'measures' package.
Usage
varImpRanger(object, data, target, nperm = 1, measure = "multiclass.Brier")
Arguments
object |
An object as returned by cforest. |
data |
Original data that was used for training the random forest. |
target |
Target variable as used in the trained model. |
nperm |
The number of permutations performed. |
measure |
The name of the measure of the 'measures' package that should be used for the variable importance calculation. |
Value
Vector with computed permutation importance for each variable.
Examples
## Not run:
library(ranger)
iris.rg = ranger(Species ~ ., data = iris, keep.inbag = TRUE, probability = TRUE)
vimp.ranger = varImpRanger(object = iris.rg, data = iris, target = "Species")
vimp.ranger
## End(Not run)