Author alan-turing-institute
1 Star
Updated Last
1 Year Ago
Started In
January 2021


Linux Coverage
Build status codecov

A light-weight package defining fall-back implementations for a collection of traits possessed by statistical objects. Here a "trait" is a function with a single argument that is a julia type, which might encode type metadata for inspection, or for use in function dispatch.

Here are two examples, which might apply to both statistical model types and loss function types:

target_scitype(::Type) = Unknown
prediction_type(::Type) = :unknown # also :probablistic, :deterministic, :interval

Here Unknown is a type defined in the light-weight package ScientificTypes.jl, the only dependency of StatsTraits.jl.

Refer to source code for a complete list.

The traits defined here are overloaded by assorted model types and measure types in the MLJ universe (examples of measures are loss functions and scoring rules). However, this small package might be of independent interest.

A planned StatisticalMeasures.jl package, to contain measures currently in MLJBase.jl, will have this package as a dependency, allowing it to be independent of the MLJ packages.