v0.20.8

MLJBase v0.20.8

Diff since v0.20.7

• (enhancement) Add support for feature importances (#798) @OkonSamuel

Merged pull requests:

• Remove apparatus to anonymize composite model fitresults (#795) (@ablaom)
• Add feature_importances machine inspection method. (#798) (@OkonSamuel)
• For a 0.20.8 release (#799) (@ablaom)

v0.20.7

MLJBase v0.20.7

Diff since v0.20.6

Closed issues:

• Handling of NaN micro/macro aggregation (#630)

Merged pull requests:

• Skip NaN in averaging in MulticlassFScore (#792) (@ablaom)
• Fix bugs and inconsistencies in documentation on evaluate!s resampling keyword (#793) (@wolthom)
• For a 0.20.7 release (#794) (@ablaom)

v0.20.6

MLJBase v0.20.6

Diff since v0.20.5

• (enhancement) Support mulithreaded training of learning networks. Call as in fit!(node, acceleration=CPUThreads()) (#785) @olivierlabayle
• (enhancement) Create interface point for specifying acceleration mode when "exporting" a learning network as new model type, by supporting acceleration as keyword argument of return! method (#785) @olivierlabayle
• (enhancement) Add acceleration and cache fields to Stack type (#785) @olivierlabayle

Merged pull requests:

• Stack cache and acceleration (rebased) (#785) (@ablaom)
• For a 0.20.6 release (#788) (@ablaom)

v0.20.5

MLJBase v0.20.5

Diff since v0.20.4

Closed issues:

• unpack error (#777)
• typename method for AbstractDataFrame incorrect equality (#784)

Merged pull requests:

• closes #784 by fixing typename + tests (#786) (@tlienart)
• For a 0.20.5 release (#787) (@ablaom)

v0.20.4

MLJBase v0.20.4

Diff since v0.20.3

• bump compat for LossFunctions.jl dependency.

Merged pull requests:

• Fix typo (#771) (@KronosTheLate)
• Bump LossFunctions.jl (#773) (@juliohm)
• bump 0.20.4 (#774) (@OkonSamuel)
• For 0.20.4 release (#776) (@OkonSamuel)

v0.20.3

MLJBase v0.20.3

Diff since v0.20.2

• Add a standard error column to the display of PerformanceEvaluation objects (as returned by evaluate!/evaluate) (#766) @rikhuijzer

Merged pull requests:

• Add std to show for PerformanceEvaluation (#766) (@rikhuijzer)
• Make running tests via TestEnv easier (#769) (@rikhuijzer)
• For a 0.20.3 release (#770) (@ablaom)

v0.20.2

MLJBase v0.20.2

Diff since v0.20.1

• (enhancement) Improve display of PerformanceEvaluation objects (output of evaluate/evaluate!) to squeeze in measures with long names and multiple parameters (#757, JuliaAI/MLJ.jl#923)
• (enhancement) Allow user to specify scitype_check_level when constructing machines, as in machine(model, X, y, check_scitype_level=2), to control how strictly the constructor enforces scitype type compatibility between model and the data. Allow user to change the global default scitype_check_level using new method scitype_check_level(i::Int).

Merged pull requests:

• Allow user to control the level of scitype checks in machine constructor (#753) (@ablaom)
• Improve short version of show for measures (#757) (@ablaom)
• For a 0.20.2 release (#763) (@ablaom)

v0.20.1

MLJBase v0.20.1

Diff since v0.20.0

Closed issues:

• For a 0.20 release (#725)
• How to transform any deterministic regression model to a probabilistic binary classification model, and then evaluate it with k-fold cross validation (#760)

Merged pull requests:

• Fix wrong field name in a show method (#758) (@rikhuijzer)
• For a 0.20.1 release (#761) (@ablaom)

v0.20.0

MLJBase v0.20.0

Diff since v0.19.8

• Relax and simplify scitype checks when constructing machines. The existing fit_data_scitype model trait encodes all allowed fit "scitype" signatures, and scitype checks now only consider this trait. In particular, an appropriately implemented transformer can now be passed a training target without tripping the type checker. (#699, #732) @pazzo83 @ablaom
• (enhancement, breaking) Redesign the serialization API to: (i) Allow use of arbitrary serialization packages for core serialization; (ii) Ensure serialization plays nicely with model composition and meta-algorithms like tuning; (iii) Ensure all traces of training data are absent in serialised models (not previously true for all composite models or if cache=true in machine constructor). Models with non-persistent learned parameters (fitresult) implement a modified model API that is documented here. The new user workflow will shortly appear in the MLJ manual under "Machines". (JuliaAI/MLJSerialization.jl#15, #733, JuliaAI/MLJSerialization.jl#16) @olivierlabayle

Closed issues:

• Relax any checks that block transformers needing to see target in training. (#699)

Merged pull requests:

• Serialization (#733) (@olivierlabayle)
• Add suggestion to err_incompatible_prediction_types message (#748) (@ablaom)
• Add a brief reminder that the fields of the PerformanceEvaluation struct are part of the public API (#749) (@DilumAluthge)
• For a 0.20 release (#751) (@ablaom)
• Fix a problem with confmat and CategoricalValue eltype (#752) (@ablaom)
• For a 0.20 release (#754) (@ablaom)
• Bump version. (#755) (@ablaom)

v0.19.8

MLJBase v0.19.8

Diff since v0.19.7

• (enhancement) Remove the restriction on resampling strategies for Stack requiring nfolds to be a field (#745) @olivierlabayle

Merged pull requests:

• make the stack's resampling scheme more general (#745) (@olivierlabayle)
• For a 0.19.8 release (#746) (@ablaom)