v0.7.3

MLJTuning v0.7.3

Diff since v0.7.2

Merged pull requests:

• Clean up tests and code to avoid private access of machine.report (#181) (@ablaom)
• For a 0.7.3 release (#182) (@ablaom)

v0.7.2

MLJTuning v0.7.2

Diff since v0.7.1

Closed issues:

• Checklist for 0.7.0 release (#152)

Merged pull requests:

• Add link to usage examples (#177) (@rikhuijzer)
• Clarify error for uninstantiated model further (#178) (@rikhuijzer)
• Update README.md (#179) (@pitmonticone)
• For a 0.7.2 release (#180) (@ablaom)

v0.7.1

MLJTuning v0.7.1

Diff since v0.7.0

• (enhancement) Allow user to specify class weights when constructing TunedModel , for passing to measures (#172)
• (convenience) Allow user to specify models as an argument instead of keyword, as in TunedModel(model, tuning=...) (#175)
• Suppress some unecessary "outer layer" caching of training data (#173)

Closed issues:

• Tuned Model interface doesnt have class_weights (#134)
• Allow TunedModel(mymodel; kwargs....) in addition to TunedModel(model=mymodel; kwargs...) (#154)
• Machines wrapping TunedModel instances should never cache data (#171)

Merged pull requests:

• Add class weight support (#172) (@ablaom)
• Suppress data caching in TunedModel machines, at outer level (#173) (@ablaom)
• Allow user to specify model in TunedModel as arg instead of kwarg (#175) (@ablaom)
• For a 0.7.1 release (#176) (@ablaom)

v0.7.0

MLJTuning v0.7.0

Diff since v0.6.16

• (breaking) Change default tuning strategy from Grid to RandomSearch() (#147)
• (breaking) Remove deprecated learning_curve! method (#151)
• (breaking) Adapt to serialization changes in MLJBase 0.20. In particular, this allows serialization of TunedModel(model=...) when model is not pure Julia (#165) @olivierlabayle

Closed issues:

• Make RandomSearch the default, instead of Grid (#147)
• Remove learning_curve! which has been deprecated (#151)

Merged pull requests:

• add serialization (#165) (@olivierlabayle)
• Fix a learning curve bug for case JULIA_NUM_THREADS = 1 (#166) (@ablaom)
• Remove deprecated learning_curve!(...) (#167) (@ablaom)
• Change tuning default to Random() instead of Grid() (#168) (@ablaom)
• Bump julia version to 1.6 in [deps] and ci (#169) (@ablaom)
• For a 0.7.0 release (#170) (@ablaom)

v0.6.16

MLJTuning v0.6.16

Diff since v0.6.15

Closed issues:

• Plotting results of tuning is not working for negative measurements (#162)

Merged pull requests:

• Fix bug to allow plotting of negative measurements (#163) (@ablaom)
• For a 0.6.16 release (#164) (@ablaom)

v0.6.15

MLJTuning v0.6.15

Diff since v0.6.14

Closed issues:

• Add warning in documentation about unpredictability of history order when using parallelization (#140)

Merged pull requests:

• Use mean instead of average for UnivariateFinite in tests; extend MLJBase compat to include 0.19^ (#160) (@ablaom)
• For a 0.6.15 release (#161) (@ablaom)

v0.6.14

MLJTuning v0.6.14

Diff since v0.6.13

Merged pull requests:

• Fix invalid test tripped by julia 1.6.3 (reproducibility issue) (#156) (@ablaom)
• Fix a confusing error when forgetting to instantiate models (#157) (@rikhuijzer)
• Enable tuning of outlier detection models (#158) (@davnn)
• For a 0.6.14 release (#159) (@ablaom)

v0.6.13

MLJTuning v0.6.13

Diff since v0.6.12

Closed issues:

• Issue for triggering new releases (#59)
• Add deprecation warning for learning_curve! (#60)
• Add particle swarm strateg(ies) (#132)
• Scores versus losses being addressed properly in TreeParzen? (#136)

Merged pull requests:

• Deprecate learning_curve! (#150) (@ablaom)
• For a 0.6.13 release (#153) (@ablaom)

v0.6.12

MLJTuning v0.6.12

Diff since v0.6.11

• (enhancement) Allow specification of deterministic measures for probabilistic predictors (#148)

Merged pull requests:

• Allow operation to be automatically inferred (#148) (@ablaom)
• For a 0.6.12 release (#149) (@ablaom)

v0.6.11

MLJTuning v0.6.11

Diff since v0.6.10

• Fix problem with reproducibility when acceleration isa CPUThreads (#126, #145) @lhnguyen-vn

Closed issues:

• Non-thread safe use of resampling machines (#126)

Merged pull requests:

• update badge ulrs in readme (#144) (@ablaom)
• Deepcopy resampler to ensure reproducibility (#145) (@lhnguyen-vn)
• For a 0.6.11 release (#146) (@ablaom)