v0.18.4

MLJBase v0.18.4

Diff since v0.18.3

Merged pull requests:

• CompatHelper: bump compat for "Distributions" to "0.25" (#546) (@github-actions[bot])
• For 0.18.4 release (#547) (@ablaom)

v0.18.3

MLJBase v0.18.3

Diff since v0.18.2

Closed issues:

• Move model serialisation out to separate package (#388)
• In resampling: Allow specification of class weights, for passing to classification metrics that support them (#457)

Merged pull requests:

• bump compat CategoricalArrays = "0.9, 0.10"; level -> levelcode (#544) (@ablaom)
• For a 0.18.3 release (#545) (@ablaom)

v0.18.2

MLJBase v0.18.2

Diff since v0.18.1

• update compatibility for PrettyTables (1.0) and Missings (1.0)

Merged pull requests:

• Update CompatHelper.yml (#537) (@DilumAluthge)
• Run CI on Julia nightly (#540) (@DilumAluthge)
• Compat updates for Missings and PrettyTable (#541) (@ablaom)
• For a 0.18.2 release (#542) (@ablaom)

v0.18.1

MLJBase v0.18.1

Diff since v0.18.0

• Suppress P <: Real check for the type of probabilities in UnivariateFinite. Allow UnivariateFinite objects to be unnormalised (ie, represent arbitrary signed measures on their discrete support). In this way also allow "probabilities" of NaN, Inf, or -Inf, to address #525 (PR #531)
• Extend compatibility requirements to support MLJModelInterface 1.0 and StatisticalTraits 1.0 (which are essentially identical to previous versions)

Closed issues:

• Some OpenML data files are materialising in format that does not support schema (#364)
• OpenML: another field name pathology? (#365)
• MLJBase + PackageCompiler + Distributed => error (#427)
• UnivariateFinite should accept NaN probabilities (#525)

Merged pull requests:

• Fixed a spelling mistake in pipeline error message (#530) (@adityasaini70)
• Allow UnivariateFinite distributions to be unnormalised, and allow NaN, Inf, -Inf (#531) (@ablaom)
• CompatHelper: bump compat for "StatisticalTraits" to "1.0" (#534) (@github-actions[bot])
• CompatHelper: bump compat for "MLJModelInterface" to "1.0" (#535) (@github-actions[bot])
• For a 0.18.1 release (#536) (@ablaom)

v0.18.0

MLJBase v0.18.0

Diff since v0.17.7

• (breaking) Remove OpenML integration, now provided by MLJOpenML

• (breaking) Remove functionality for saving and retrieving machines, now provided by MLJSerialization

Merged pull requests:

• Remove code now moved to MLJSerialization and MLJOpenML (#526) (@giordano)
• For a 0.18.0 release (#527) (@ablaom)
• Update readme (#528) (@ablaom)

v0.17.7

MLJBase v0.17.7

Diff since v0.17.6

• Add support for iterative models (#523, JuliaAI/MLJ.jl#139)

Merged pull requests:

• Add support for iteration (#523) (@ablaom)
• For a 0.17.7 release (#524) (@ablaom)

v0.17.6

MLJBase v0.17.6

Diff since v0.17.5

Merged pull requests:

• Relax dimension check in machine constructor (#521) (@ablaom)
• For a 0.17.6 release (#522) (@ablaom)

v0.17.5

MLJBase v0.17.5

Diff since v0.17.4

• (enhancement) Allow user to specify class _weights=... in evaluate/evaluate!, in the same way that (per-sample) weights are currently passed. This will pass the class weights to measures that support them. These include: MulticlassFScore, MulticlassFalseDiscoveryRate, MulticlassFalseNegativeRate, MulticlassFalsePositiveRate, MulticlassNegativePredictiveValue, MulticlassPrecision, MulticlassTrueNegativeRate, and MulticlassTruePositiveRate (#515) @ven-k @ablaom

Merged pull requests:

• Add class_weights=... in evaluate! (#515) (@ablaom)
• For a 0.17.4 release (#519) (@ablaom)
• For a 0.14.5 release (take 2) (#520) (@ablaom)

v0.17.4

MLJBase v0.17.4

Diff since v0.17.3

Merged pull requests:

• get rid of dep warnings associated with use of CategoricalArrays.get (#517) (@ablaom)
• For a 0.17.4 release (#518) (@ablaom)

v0.17.3

MLJBase v0.17.3

Diff since v0.17.2

Merged pull requests:

• For a 0.17.3 release (#516) (@ablaom)