I am pleased to announce two new packages for measures (aka metrics) for statistics and machine learning:
Build complex measures with StatisticalMeasuresBase.jl
Use StatisticalMeasuresBase.jl to quickly build production-ready measures out of simpler functions. See here for the basic idea.
Or grab a ready-to-use measure from StatisticalMeasures.jl
StatisticalMeasures.jl provides a large library of measures satisfying the simple API defined in StatisticalMeasuresBase.jl. These measures are ready to use with the popular machine learning framework MLJ.jl (version ≥ 0.12) which previously provided many of the measures natively. (MLJ users, see also this migration guide.) Measures include confusion matrices, new multi-target measures for time series. A receiver operator characteristic tool is included.
History and acknowledgments. These packages grew out of the library of measures previously part of MLJBase.jl. We created the new packages to encourage wider application, and to make growing the library easier. While the API has been redesigned, the core functionality can be considered well-tested and mature.
I acknowledge contributions to the original code base from the following people (GitHub handles): Thibaut Lienart (@tlienart), Venkateshprasad (@ven-k), Samuel Okon (@OkonSamuel), Rik
Huijzer (@rikhuijzer), Mose Giordano (@giordano), Cameron Biegnek (@CameronBieganek),
David Muhr (@davnn), Albert Alex Zevelev (@azev).
I’ve been wanting a lightweight package like this with only statistical measures. Great work!