v0.4.1

MLJIteration v0.4.1

Diff since v0.4.0

Closed issues:

• Catch models that do not overload training_losses, reporting informative error message (#35)
• InvalidValue is not triggering a stop for invalid training losses (#36)

Merged pull requests:

• Fix tests for change in |> in MLJBase 0.19 (backwards compatible fix) (#37) (@ablaom)
• CompatHelper: bump compat for MLJBase to 0.19, (keep existing compat) (#38) (@github-actions[bot])
• For a 0.4.1 release (#39) (@ablaom)

v0.4.0

MLJIteration v0.4.0

Diff since v0.3.3

• (breaking) Bump the compatibility requirement for IterationControl.jl, which changes slightly the behaviour of the PQ stopping criterion. See these release notes.

Merged pull requests:

• CompatHelper: bump compat for IterationControl to 0.5, (keep existing compat) (#33) (@github-actions[bot])
• For a 0.3 release (#34) (@ablaom)

v0.3.3

MLJIteration v0.3.3

Diff since v0.3.2

• Add WithModelDo (#30) @ablaom

Merged pull requests:

• Add WithModelDo (#30) (@ablaom)
• For a 0.3.3 release (#31) (@ablaom)

v0.3.2

MLJIteration v0.3.2

Diff since v0.3.1

• Add new controls WithFittedParamsDo, WithReportDo, WithMachineDo.

Closed issues:

• Issue to trigger new releases (#2)
• Add a WithFittedParametersDo control (#25)
• Add a WithReportDo control strategy (#27)

Merged pull requests:

• Fix a test using ScientificTypes instead of ScientificTypesBase (#26) (@ablaom)
• Add the controls WithFittedParamsDo, WithReportDo, WithMachineDo (#28) (@ablaom)
• For a 0.3.2 release (#29) (@ablaom)

v0.3.1

MLJIteration v0.3.1

Diff since v0.3.0

• Remove warning about slow down for Holdout(rng=...) as no longer applicable. Allow user to explicitly specify a train/test holdout set (of row indices), as in resampling = [(train, test), ] for some integer vectors train and test (as in evaluate!) (#12, #22)

Merged pull requests:

• Remove warnings when resampling=Holdout(rng=...) (#22) (@ablaom)
• For a 0.3.1 release (#23) (@ablaom)

v0.3.0

MLJIteration v0.3.0

Diff since v0.2.3

• Update control API to IterationControl 0.4, which has a new signature for IterationControl.update!(control, ...).

Merged pull requests:

• Bump IterationControl requirement to 0.4 (#20) (@ablaom)
• For a 0.3.0 release (#21) (@ablaom)

v0.2.3

MLJIteration v0.2.3

Diff since v0.2.2

Closed issues:

• IteratedModel not training after IterationControl update (#17)

Merged pull requests:

• Logging test update (no new release) (#16) (@ablaom)
• Training losses bug fix (#18) (@ablaom)
• For a 0.2.3 release (#19) (@ablaom)

v0.2.2

MLJIteration v0.2.2

Diff since v0.2.1

• Some more informative logging (#11, #13)

Closed issues:

• Make it clearer in docs that one control must be a training control (#11)
• Issue warning if shuffling the Holdout sample (#12)

Merged pull requests:

• Add warning for use of shuffling in holdout (#13) (@ablaom)
• Catch controls that don't include training (#14) (@ablaom)
• For a 0.2.2 release (#15) (@ablaom)

v0.2.1

MLJIteration v0.2.1

Diff since v0.2.0

Merged pull requests:

• Stop statement about resetting measure when resampling=nothing (#9) (@ablaom)
• For a 0.2.1 release (#10) (@ablaom)

v0.2.0

MLJIteration v0.2.0

Diff since v0.1.1

• (breaking) Change default value of f in WithIterationsDo(f=...) to f(i) = @info("num iterations: $i")
• (breaking) Remove Save control (now moved to MLJSerialization)

Merged pull requests:

• Change WithIterationsDo default value of f (#6) (@ablaom)
• Remove Save control and bump MLJBase compat to 0.18 (#7) (@ablaom)
• For a 0.2.0 release (again) (#8) (@ablaom)