v0.18.24

MLJBase v0.18.24

Diff since v0.18.23

• Implement default measures for outlier detectors (#662) @davnn

Merged pull requests:

• Typo in TimeSeriesCV (#661) (@ParadaCarleton)
• Improvements for detectors and removal of unnecessary exports (#662) (@davnn)
• For a 0.18.24 release (#663) (@ablaom)

v0.18.23

MLJBase v0.18.23

Diff since v0.18.22

• Generalize FScore so as to handle some corner cases (#650)
• Add adjusted option to BalancedAccuracy (#569)
• Make some improvements to show methods (#532, #654, #273)
• Allow predict_mode to work with Unsupervised models (#658)
• Fix input checks for UnsupervisedAnnotator (and so for unsupervised outlier detector models) (#657)

Closed issues:

• Add more informative show for samplers (#273)
• Improve show for one-dimensional range objects (#532)
• Add adjusted option to BalancedAccuracy (#569)
• Balanced accuracy comparison with sk-learn (#651)

Merged pull requests:

• Fix binary FScore corner case (#650) (@ablaom)
• Add adjusted balanced accuracy (#652) (@ablaom)
• Revert balanced accuracy testing (#653) (@ablaom)
• Display tweaks (#654) (@ablaom)
• Improve sampler object display (#655) (@ablaom)
• Fix input checks for unsupervised annotators (#657) (@davnn)
• Extend overloading of predict_mode, etc to all models, not just Supervised (#658) (@ablaom)
• For a 0.18.23 release (#659) (@ablaom)

v0.18.22

MLJBase v0.18.22

Diff since v0.18.21

• Fix bug with classification metrics exposed by Julia 1.6.3 release (#646) @OkonSamuel
• Remove some corner cases of NaN in MulticlassFScore that are resolved by eliminating common TP factor in defining quotient (#637) @OkonSamuel

Closed issues:

• @mlj_model macro fails on negative default args [needs docs for way around] (#68)
• Make all measures return a vector given vector arguments, even AUC, and so forth (#308)
• Slow machine construction for large number of features (#428)
• Fix documentation fail (#613)
• Fix show for ranges (#641)

Merged pull requests:

• Relax machine argument type check slightly (#636) (@ablaom)
• Use expanded formula of MulticlassFScore (#637) (@ven-k)
• Fix typo (#640) (@ParadaCarleton)
• Reconnect some orphaned tests (#645) (@ablaom)
• fix bug in _confmat (#646) (@OkonSamuel)
• For a 0.18.22 release (#647) (@ablaom)
• Remove TypedTables as dependency (it is only a test dependency). To fix failed release (#649) (@ablaom)

v0.18.21

MLJBase v0.18.21

Diff since v0.18.20

• (bug fix) Correct bug in MulticlassTrueNegative (#629)

Merged pull requests:

• Multiclass true negative of the conf_mat fix :) (#629) (@filippfarias)
• For a 0.18.21 release (#634) (@ablaom)

v0.18.20

MLJBase v0.18.20

Diff since v0.18.19

• (enhancement) Integrate new outlier detection types added to MLJModelInterface (JuliaAI/MLJ.jl#780, JuliaAI/MLJModelInterface.jl#113) and adapt performance evaluation methods to handle them (#628)

Relating to measures:

• (enhancement) Measures: Add SphericalScore and LogScore (negative of LogLoss).
• (enhancement) Extend the proper scoring rules SphericalScore, LogScore and BrierScore to handle Continuous and Count data. Supported distributions types from Distributions are: Uniform, Normal, Exponential, Logistic, Chi, Chisq, Beta, Gamma, Cauchy, Poisson, DiscreteUniform, DiscreteNonParameteric (#627)
• (enhancement) Add missing and NaN support for all measures, excluding AreaUnderCurve and measures from LossFunctions.jl (which imported library does not support) (#616)
• (enhancement) Add skipinvalid(y) and skipinvalid(yhat, y) methods. The first returns an iterator that skips missing and NaN values - similar to skipmissing and is performant. The second returns the flattened forms of yhat and y with invalid entries removed "commensurately" from both yhat and y, meaning an element of either argument is skipped even if valid, if the corresponding element of the other argument is invalid (#627). This method is necessarily less efficient and provided for convenience for pre-processing data for the measures which do not support invalid entries (#627, #616)
• (enhancement) Allow most measures for Finite data to be called with "raw" data, that is, data that is not wrapped as CategoricalArray. This includes ConfusionMatrix. A warning is issued to indicate order ambiguity, with the usual suggestion to coerce to OrderedFactor to suppress the warning (#627)
• (enhancement) Allow measures to be called on arrays, and not just vectors (#627) but see remaining limitations at #631.
• (API) Make implementing new measures simpler (#627) and less error-prone. See this guide for details.
• (enhancement) Introduce new method MLJBase.call(measure, args...) to call a measure without applying dimension or pool checks.
• (bug fix) Prevent weights passed to measures from Loss functions.jl from being normalized (#626)

Closed issues:

• Add BrierScore for Continuous targets and assorted Distributions (#395)
• Add support for missing/NaN values where possible in measures (#616)
• Improve the doc-string for unpack (#620)
• Skip NaN in aggregation of measures (#622)
• Weights passed to measures from LossFunctions are being normalized (#626)

Merged pull requests:

• Add Continuous and Count versions of Brier score (#623) (@ablaom)
• Skip NaN as well as missing in measure aggregation (#624) (@ablaom)
• Add Infinite versions of the proper scoring rules spherical, log (#625) (@ablaom)
• Big measures cleanup (#627) (@ablaom)
• Outlier Detection Integration (#628) (@davnn)
• For a 0.18.20 release (#632) (@ablaom)
• Last minute bump of Distributions compat (#633) (@ablaom)

v0.18.19

MLJBase v0.18.19

Diff since v0.18.18

• (enhancement) Extend Finite measures to accept missing values (#618)
• (enhancement) In cases of a probabilistic model, have evaluate! automatically choose the operation (predict_mode, predict_mean, etc) when user specifies a deterministic measure. Allow a mixture of deterministic and probabilistic measures to be specified for such models. Allow automatic behaviour to be explicitly overridden by specifying a vector operations=.... This mitigates a common gotcha for new users (#598, #599, #600)

Closed issues:

• Have evaluate! automatically run the right kinds of predictions for each metric (#598)
• Add a wrapper for measures for use with missing values? (#602)

Merged pull requests:

• Alternative to #599 with less logging (#600) (@ablaom)
• Implement pdf(::UnivariateFinite, ::Missing) = missing and cousins (#617) (@ablaom)
• Allow finite measures, incl confmat, to handle missing values (#618) (@ablaom)
• For a 0.18.19 release (#619) (@ablaom)

v0.18.18

MLJBase v0.18.18

Diff since v0.18.17

Closed issues:

• Issue to trigger releases (#345)

Merged pull requests:

• Cleanup some error/warning code for machine constructors (#607) (@ablaom)
• Add better scitype check for machine constructor when model is not standard (#611) (@ablaom)
• For a 0.18.18 release (#612) (@ablaom)

v0.18.17

MLJBase v0.18.17

Diff since v0.18.16

Merged pull requests:

• Abstract type cleanup (#605) (@ablaom)
• For a 0.18.17 release (#606) (@ablaom)

v0.18.16

MLJBase v0.18.16

Diff since v0.18.15

Merged pull requests:

• Fix typo for unpack documentation (unnpack) (#601) (@cmey)
• Fix show for some LossFunctions measures (#603) (@ablaom)
• For a 0.18.16 release (#604) (@ablaom)

v0.18.15

MLJBase v0.18.15

Diff since v0.18.14

Closed issues:

• Review of MLBase and things we could port over (#94)

Merged pull requests:

• add docstring fix (#593) (@olivierlabayle)
• Fix typo in test/resampling.jl (#594) (@rikhuijzer)
• Remove model type parameter from Resampler (#596) (@ablaom)
• For a 0.18.15 release (#597) (@ablaom)