v0.1.14

MLJFlux v0.1.14

Diff since v0.1.13

Merged pull requests:

• Replace Flux.Data with MLDatasets (#141) (@ayush-1506)
• Bump version (#143) (@ablaom)
• For a 0.1.14 release (#144) (@ablaom)
• mv MLDatasets to [extras] (#145) (@ablaom)

v0.1.13

MLJFlux v0.1.13

Diff since v0.1.12

Merged pull requests:

• Small typo in paper (#133) (@ayush-1506)
• rm LossFunctions from [deps] (#138) (@ablaom)

v0.1.12

MLJFlux v0.1.12

Diff since v0.1.11

• Remove type piracy for optimisers (#135)
• Fix incorrect cold-start behaviour (#124)
• Make builder and optimiser "deep" hyper-parameters (#135)

Closed issues:

• Cold training restart where warm restart is wanted (#124)
• Fix type piracy for optimisers (#129)

Merged pull requests:

• Remove optimiser type piracy. Make optimiser and builder "deep" hyper-parmaters (#135) (@ablaom)

v0.1.11

MLJFlux v0.1.11

Diff since v0.1.10

Merged pull requests:

• paper review (#123) (@ayush-1506)
• CompatHelper: bump compat for "CategoricalArrays" to "0.10" (#128) (@github-actions[bot])
• CompatHelper: bump compat for "ColorTypes" to "0.11" (#130) (@github-actions[bot])
• Update Project.toml (#131) (@ablaom)
• for a 0.1.11 release (#132) (@ablaom)

v0.1.10

MLJFlux v0.1.10

Diff since v0.1.9

• extend compat for Flux to include 0.12
• extend compat for MLJModelInterface to include 1.0

Closed issues:

• Switch code coverage to use Codecov? (#82)
• Specifying non-supported acceleration option should trigger fallback to CPU1() (#104)
• Add support for Flux version 0.12? (#107)

Merged pull requests:

• Add acceleration cleanup in clean! (#105) (@ayush-1506)
• CompatHelper: bump compat for "Flux" to "0.12" (#110) (@github-actions[bot])
• Revert "CompatHelper: bump compat for "Flux" to "0.12"" (#111) (@DilumAluthge)
• CompatHelper: bump compat for "Flux" to "0.12" (#116) (@github-actions[bot])
• Update the MNIST example to make use of IteratedModel wrapper (#125) (@ablaom)
• Extend [compat] for MLJModelnterface to include 1.0, and bump MLJFlux version (#126) (@ablaom)
• For a 0.1.10 release (#127) (@ablaom)

v0.1.9

MLJFlux v0.1.9

Diff since v0.1.8

• Add support for MLJ's forthcoming iteration interface (#100, JuliaAI/MLJ.jl#139)

Merged pull requests:

• Add .gitattributes (#98) (@ayush-1506)
• Add .gitattributes (#99) (@ablaom)
• Add support for MLJ iteration API (#100) (@ablaom)

v0.1.8

MLJFlux v0.1.8

Diff since v0.1.7

Closed issues:

• Update examples/ or remove to new dir outdated/ (#56)
• Add a GPU example to the README? (#78)
• Replace train! with a custom loop (#89)
• Suppress gpu testing in PR to dev branch? (#92)

Merged pull requests:

• Examples review (#90) (@ablaom)
• Doc improvements. No new release (#91) (@ablaom)
• Only run Buildkite (GPU CI) on PRs to master (#93) (@DilumAluthge)
• Custom training loop (#95) (@ayush-1506)
• For a 0.1.8 release (#96) (@ablaom)

v0.1.7

MLJFlux v0.1.7

Diff since v0.1.6

Closed issues:

• CompatHelper failing (#55)
• TODO: Set up the GPU continuous integration (CI) for MLJFlux.jl (#65)

Merged pull requests:

• Add license (#80) (@ablaom)
• Set up Buildkite (#81) (@DilumAluthge)
• Remove a broken link from the README (#83) (@DilumAluthge)
• Tests: relax tolerance for gpu/cpu agreement in image case (#84) (@ablaom)
• Carry over optimiser state in update() methods (#86) (@ablaom)
• For a 0.1.7 release (#88) (@ablaom)

v0.1.6

MLJFlux v0.1.6

Diff since v0.1.5

Merged pull requests:

• For a 0.1.6 release (#76) (@ablaom)
• For a 0.1.6 release (#77) (@ablaom)

v0.1.5

MLJFlux v0.1.5

Diff since v0.1.4

Merged pull requests:

• Move from Travis CI to GitHub Actions CI (#72) (@DilumAluthge)
• switch ci to gh action & extend [compat] CategoricalArrays="...,0.9" (#73) (@ablaom)
• Extend [compat] Flux = "...,0.11" (#74) (@ablaom)
• For a 0.1.5 release (#75) (@ablaom)