v0.18.14

MLJBase v0.18.14

Diff since v0.18.13

• Improve performance of nrows for row-based tables (#591) @OkonSamuel

Merged pull requests:

• improve performance of MMI.nrows (#591) (@OkonSamuel)
• For a 0.18.14 release (#592) (@ablaom)

v0.18.13

MLJBase v0.18.13

Diff since v0.18.12

• (enhancent) Add CVTimeSeries resampling strategy, for time series data (#331) @aa25desh @CameronBieganek

Merged pull requests:

• Continuation of "adds cross-validation for time series" #278 (#331) (@ablaom)
• For a 0.18.13 release (#590) (@ablaom)

v0.18.12

MLJBase v0.18.12

Diff since v0.18.11

Closed issues:

• New patch release of LossFunctions is triggering build fail (#575)

Merged pull requests:

• Fix some formatting in stack doc-string and simplify a bit (#583) (@ablaom)
• Update README.md (#585) (@OkonSamuel)
• Update doc deployment url (#587) (@ablaom)
• Address ScientificTypes name change (#588) (@ablaom)
• For a 0.18.12 release (#589) (@ablaom)

v0.18.11

MLJBase v0.18.11

Diff since v0.18.10

Merged pull requests:

• Replacing length by nrows to solve #580 (#581) (@Leonardbcm)

v0.18.10

MLJBase v0.18.10

Diff since v0.18.9

• (ehnancement) Add canned version of Wolpert model stacking, Stack (#564) @olivierlabayle @ablaom
• (bug fix) Fix overflow bug in MatthewsCorrelation coefficient (#571) @giordano
• (enhancement) Overload partition to work on arbitrary vectors, matrices and tables (#361) @giordano
• (bug fix) Fix flawed interface point for RNG specification in synthetic data methods, such as make_blobs (#567) @OkonSamuel
• Remove LossFunctions.jl type piracy (#577) @ablaom

Closed issues:

• Review of loading time (#77)
• Generalize partition to split any table or matrix or vector, not just vector of integers (#361)
• Use xlogy function in cross-entropy definition (#373)
• Balanced accuracy on ConfusionMatrix fails (#380)
• Matthews correlation should be reviewed (#381)
• Cannot pass arbitrary rng to make_blobs (#567)

Merged pull requests:

• Add method partition(::AbstractMatrix, ...) (#561) (@giordano)
• Stacking (#564) (@olivierlabayle)
• Tighten return type of predict for KNNClassifier (model for testing) (#566) (@ablaom)
• address issue 567 (#568) (@OkonSamuel)
• Fix computation of Matthews correlation coefficient (#571) (@giordano)
• Remove LossFunctions type piracy (#577) (@ablaom)
• Fix transposed y and yhat in measures from LossFunctions (#578) (@ablaom)

v0.18.9

MLJBase v0.18.9

Diff since v0.18.8

Merged pull requests:

• Bump compat ProgressMeter = "1.7.1" and version = "0.18.9" (#562) (@ablaom)

v0.18.8

MLJBase v0.18.8

Diff since v0.18.7

• Fix bug with Resampler (#559)
• Add train_test_rows field to performance evaluation objects (#559)
• Improve error message for attempts to call report/fitted_params on an untrained machine (#558)

Merged pull requests:

• Add more informative error for report/fitted_param on untrained mach (#558) (@ablaom)
• Ensure Resampler does not generate new train/test pairs unless resampling or repeats changes (#559) (@ablaom)

v0.18.7

MLJBase v0.18.7

Diff since v0.18.6

Merged pull requests:

• CompatHelper: bump compat for "LossFunctions" to "0.7" (#555) (@github-actions[bot])
• Fix error message for using deterministic measure with probabilistic model in evaluate! (#556) (@ablaom)

v0.18.6

MLJBase v0.18.6

Diff since v0.18.5

• Make the show method, and fitted_params/report for composite models, property-based, instead of field-based (#553 )

Merged pull requests:

• Update resampling.jl (#552) (@adityasaini70)
• Make MLJType show method, and fitted_params/report for composites, property-based instead of field-based (#553) (@ablaom)

v0.18.5

MLJBase v0.18.5

Diff since v0.18.4

Closed issues:

• unexpected behaviour for fitted_params when a composite model has == component models (#549)

Merged pull requests:

• Resolve #549 (#550) (@ablaom)
• For a 0.18.5 release (#551) (@ablaom)