v0.12.4

MLJBase v0.12.4

Diff since v0.12.3

Bump [compat] Distributions="^0.22,^0.23"

Merged pull requests:

• For a 0.12.4 release (#225) (@ablaom)

v0.12.3

MLJBase v0.12.3

Diff since v0.12.2

Merged pull requests:

• Documentation update for github pages (#223) (@Arkoniak)
• Auto docs (#224) (@tlienart)

v0.12.2

MLJBase v0.12.2

Diff since v0.12.1

Minor update adding methods for samplers to the scale function (currently implemented for one-dimensional ranges only).

Merged pull requests:

• For 0.12.2 release (#221) (@ablaom)

v0.12.1

MLJBase v0.12.1

Diff since v0.12.0

(enhancement) Add serialization for machines. Serialization is model-specific with a fallback implementation using JLSO. The user serializes with MLJBase.save and de-serializes with machine. See the MLJBase.save doc-string for details. Model-specific serialization is defined by methods MLJBase.save(model::MyModel, ...) and de-serialization with MLJBase.restore(model::MyModel, ...) (PR #218, JuliaAI/MLJ.jl#138, JuliaAI/MLJ.jl#292)

Merged pull requests:

• Add serialization (#218) (@ablaom)
• For 0.12.1 release (#219) (@ablaom)

v0.12.0

MLJBase v0.12.0

Diff since v0.11.10

• (enhancement) Add sampler wrapper for one-dimensional ranges, for random sampling from ranges using rand (#213)

• (breaking) Suppress normalisation of measure weights (PR #208)

• (breaking) Migrate equality.jl to MLJModelInterface (#178)

• (breaking) Shift the optional rng argument of iterator to first position (#215)

Merged pull requests:

• For a 0.11.10 release (#210) (@ablaom)
• Add sampler method for one-dimensional ranges (#213) (@ablaom)
• For a 0.12.0 release (#216) (@ablaom)
• For 0.12.0 release (#217) (@ablaom)

v0.11.10

MLJBase v0.11.10

Diff since v0.11.9

• (bug) Resolve machine constructor ambiguity for Static transformers, preventing their reliable use in composite models.

• (bug) Resolve StackOverflow error when trying to export a learning network to a composite model type with no hyperparameters, as in @from_network MyComp() <= yhat.

Merged pull requests:

• for 0.11.9 release (#206) (@ablaom)
• adding 1.0 in the CI (#207) (@tlienart)
• Bug: fix method ambiguity for machine(::Static, args...) (#209) (@ablaom)
• For a 0.11.10 release (#210) (@ablaom)

v0.11.9

MLJBase v0.11.9

Diff since v0.11.8

Bump [compat] Tables = "^1.0"

Closed issues:

• Error with evaluate! tests in the distributed setting (#169)

Merged pull requests:

• For a 0.11.8 release (#202) (@ablaom)
• compatible with MLJSt 0.2 and Tables 1 (#204) (@tlienart)
• for 0.11.9 release (#206) (@ablaom)

v0.11.8

MLJBase v0.11.8

Diff since v0.11.7

• (bug) Remove multithreading race conditions in resampling (use of evaluate!) (#169)

• (testing) Fix and re-instate all CPUThreads and CPUProcesses tests in resampling

• (enhancement) Add progress meter to all forms of accelerated resampling

• (testing) Increase number of workers in travis testing to 2 (3 processes); increase number of threads to 2.

Closed issues:

• Add shuffle=false, rng=GLOBAL_RNG keywords to unpack command (#167)

Merged pull requests:

• Fix resampling race condition, and so forth (#201) (@ablaom)
• For a 0.11.8 release (#202) (@ablaom)

v0.11.7

MLJBase v0.11.7

Diff since v0.11.6

Closed issues:

• Confusion matrices as measures (#168)

Merged pull requests:

• Make confusion_matrix into a measure (orientation = :other). For 0.11.6 release (#196) (@ablaom)
• Bringing predict_median back (#197) (@tlienart)
• Bringing predict_median back (#197) (#198) (@tlienart)

v0.11.6

MLJBase v0.11.6

Diff since v0.11.5

Allow confusion_matrix (or confmat) to be specified as a measure in calls to evaluate, evaluate! (#168, PR #195)

Merged pull requests:

• update readme (#192) (@ablaom)
• Make confusion_matrix into a measure (#195) (@ablaom)