Perform a posthoc Friedman-Nemenyi test.

Source: R/friedmanPostHocTestBMR.R

Performs a PMCMRplus::frdAllPairsNemenyiTest for a BenchmarkResult and a selected measure.

This means all pairwise comparisons of learners are performed. The null hypothesis of the post hoc test is that each pair of learners is equal. If the null hypothesis of the included ad hoc stats::friedman.test can be rejected an object of class pairwise.htest is returned. If not, the function returns the corresponding friedman.test.

Note that benchmark results for at least two learners on at least two tasks are required.

friedmanPostHocTestBMR(
  bmr,
  measure = NULL,
  p.value = 0.05,
  aggregation = "default"
)

Arguments

bmr

(BenchmarkResult)
Benchmark result.
measure

(Measure)
Performance measure. Default is the first measure used in the benchmark experiment.
p.value

(numeric(1))
p-value for the tests. Default: 0.05
aggregation

(character(1))
“mean” or “default”. See getBMRAggrPerformances for details on “default”.

Value

(pairwise.htest): See PMCMRplus::frdAllPairsNemenyiTest for details. Additionally two components are added to the list:

  • f.rejnull (logical(1)):
    Whether the according friedman.test rejects the Null hypothesis at the selected p.value

  • crit.difference (list(2)):
    Minimal difference the mean ranks of two learners need to have in order to be significantly different

See also

Other benchmark: BenchmarkResult, batchmark(), benchmark(), convertBMRToRankMatrix(), friedmanTestBMR(), generateCritDifferencesData(), getBMRAggrPerformances(), getBMRFeatSelResults(), getBMRFilteredFeatures(), getBMRLearnerIds(), getBMRLearnerShortNames(), getBMRLearners(), getBMRMeasureIds(), getBMRMeasures(), getBMRModels(), getBMRPerformances(), getBMRPredictions(), getBMRTaskDescs(), getBMRTaskIds(), getBMRTuneResults(), plotBMRBoxplots(), plotBMRRanksAsBarChart(), plotBMRSummary(), plotCritDifferences(), reduceBatchmarkResults()

Examples

# see benchmark