v0.4.0

MLJTuning v0.4.0

Diff since v0.3.6

Note. This release does not change the API for tuning strategies.

Closed issues:

• Remove unsafe use of dictionaries in multithreaded version of tuning (#48)

Merged pull requests:

• remove unsafe use of dictionary in multithreading (#49) (@OkonSamuel)
• For a 0.4.0 release (#56) (@ablaom)
• Conconrency (#57) (@OkonSamuel)
• MLJBase 0.14 update (#58) (@ablaom)

v0.3.6

MLJTuning v0.3.6

Diff since v0.3.5

Merged pull requests:

• Update models for testing to fix CI (#47) (@ablaom)
• Extend [compat] MLJModelInterface = "^0.2,^0.3" (#50) (@ablaom)
• For a 0.3.6 release (#52) (@ablaom)

v0.3.5

MLJTuning v0.3.5

Diff since v0.3.4

• Extend [compat] RecipesBase = "^0.8,^0.9,^1.0" (#45)

v0.3.4

MLJTuning v0.3.4

Diff since v0.3.3

• (enhancement) Re-instate progress meters for tuning and learning curves (#14, PR #42) (@OkonSamuel)
• (performance) Enable multithreading in tuning (acceleration=CPUThreads()) (#15, PR #42) (@OkonSamuel)
• Improve error message reporting around accelerationoptions for tuning (#27, #39, PR #42) (@OkonSamuel)
• Fixed broken links in readme (#37) (@OkonSamuel)

v0.3.3

MLJTuning v0.3.3

Diff since v0.3.2

Extend [compat] for MLJBase to "^0.12,^0.13"

Closed issues:

• learning_curve is not using "smart" fitting (performance issue) (#34)

Merged pull requests:

• Extend [compat] for MLJBase for a 0.3.3 release (#38) (@ablaom)

v0.3.2

MLJTuning v0.3.2

Diff since v0.3.1

• In Grid search, allow multiple one-dimensional ranges in a range vector to share the same field attribute, with the effect that their grids are combined. Allows one to add specific values to automatically generated ones, for example

• Improve docs on subject of implementing Cartesian range objects (#24)

Closed issues:

• Add recommendation to implement built-in Cartesian ranges in new tuning strategies (#24)
• For a 0.3.1 release (#28)

Merged pull requests:

• For a 0.3.2 release (#32) (@ablaom)
• For a 0.3.2 release (#33) (@ablaom)

v0.3.1

MLJTuning v0.3.1

Diff since v0.3.0

• (enhancement) Add RandomSearch tuning strategy (JuliaAI/MLJ.jl#37 PR #30)

Closed issues:

• For a 0.3 release (#22)

Merged pull requests:

• Implement RandomSearch (#30) (@ablaom)
• For a 0.3.1 release (#31) (@ablaom)

v0.3.0

MLJTuning v0.3.0

Diff since v0.2.0

• (breaking) Add enhancements the tuning strategy interface. This should not break use of TunedModel, and the built-in strategies Grid and Explicit are updated with no new behaviour (#21)

• (breaking) Add n_remaining argument to models! method to give access to number of evaluations remaining. This should not break use of TunedModel, and the built-in strategies Grid and Explicit are updated with no new behaviour (#23 PR #26)

Closed issues:

• Add n_remaining argument to models! (breaking) (#23)

Merged pull requests:

• Enhancements to the tuning interface (#21) (@ablaom)
• Add n_remaining argument to models! (#26) (@ablaom)
• For a 0.3 release (#29) (@ablaom)

v0.2.0

MLJTuning v0.2.0

Diff since v0.1.4

• (breaking) Update to MLJBase 0.12.0, which changes the position of the optional argument rng in iterator to first position (JuliaAI/MLJBase.jl#215)

Merged pull requests:

• For a 0.1.4 release (#17) (@ablaom)
• Update to MLJBase 0.12.0 (#19) (@ablaom)
• For 0.2.0 release (#20) (@ablaom)

v0.1.4

MLJTuning v0.1.4

Diff since v0.1.3

Update [compat] Tables = "^1.0"

Merged pull requests:

• Update to MLJBase 0.11.1 (#11) (@ablaom)
• Fixed typo in README.md (#12) (@iqml)
• fixed typo in Grid docstring (#13) (@OkonSamuel)
• bumping up compats, now ok with Tables v1 (#16) (@tlienart)
• For a 0.1.4 release (#17) (@ablaom)