- Added
FilterBoruta
- Fixed issue with
FilterPerformance
where the argmeasure
wasn't passed on - Added
FilterUnivariateCox
(thanks to @bblodfon) - Parameter value
na.rm
is properly initialized toTRUE
(thanks to @bblodfon) - Bugfix: property
missings
is now set correctly forFilterFindCorrelation
- Bugfix:
$hash
now works forFilter
s
- Tagged multiple filters to be able of gracefully handling missing values.
- Added more supported feature types to FilterCarScore.
- Improved documentation.
- Features are now checked for missing values to improve error messages (#140)
- Removed deprecated functions
- Use featureless learner in defaults (#124)
- Field
task_type
of classFilter
has been renamed totask_types
.
- Add
FilterCarSurvScore
(#120, @mllg) - Use featureless learner instead of rpart as default learner for
FilterImportance
andFilterPerformance
(#124) - Add documentation for PipeOpFilter
- Add mlr3pipelines examples to help pages (#135, @sebffischer)
- Add
label
arg toFilter
class (#121, @mllg)
- Add references to benchmark paper and praznik paper (#104)
- New filter
FilterSelectedFeatures
which makes use of embedded feature selection methods of learners. See the help page for more details (#102) - Allow
NA
as task type. This makes it possible to use other tasks than"regr"
or"classif"
for certain filters, e.g.FilterVariance
(#106)
- Fixes an issue where argument
nfeat
was not passed down to {praznik} filters (#97)
- Disable threading in praznik filters by default (5f24742e9b92f6a5f828c4f755be3fb53427afdb, @mllg)
Enable by setting hyperparameter
threads
>= 2 or to0
for auto-detection of available cores (#93, @mllg) - Document return type of private
.calculate()
(#92, @mllg) - Allow
NA
in returned vectors. Features with missing values as well as features with no calculated score are automatically ranked last, in a random order. (#92, @mllg) - praznik filters now also support
regr
Tasks (#90, @bommert)
- Add ReliefF filter (#86)
- Fix praznik scores calculation: praznik filters are not monotone in the selected features due to their iterative fashion. E.g., the first selected feature can have a score of 5, the second selected feature a score of 10. This version replaces the praznik scores by a simple sequence (#87, @mllg)
- Add Permutation (#70)
- Add
flts()
(#77) - Github Actions: set cron job to 4am to avoid potential download issues with R-devel on macOS
- Filters now have a help method
$help()
which opens the respective help page (#68)
- Use
private$.calculate
instead of public "calculate" method for Filters - switch from Travis to GitHub Actions
- Use Roxygen R6 notation for docs
- new filter
FilterFindCorrelation
(#62, @mb706)
- Replace dependency
Metrics
withmlr3measures
.
- Initial CRAN release.