v0.2.6

MLJFlux v0.2.6

Diff since v0.2.5

Closed issues:

• Ensure use of StableRNGs in all testing. (#195)

Merged pull requests:

• Fix an improper test triggering a Flux.Losses warning (#191) (@ablaom)
• Tidy up (#192) (@ablaom)
• Fix docstring (#193) (@jbrea)
• fix MLP (#194) (@jbrea)
• Use StableRNGs more comprehensively. (#196) (@ablaom)
• For a 0.2.6 release (#197) (@ablaom)

v0.2.5

MLJFlux v0.2.5

Diff since v0.2.4

• Add builder for multilayer perceptron, MLJFlux.MLP (#188 #190) @john-waczak

Closed issues:

• Issue to trigger new releases (#47)

Merged pull requests:

• added new builder for sklearn style mlp (#188) (@john-waczak)
• For a 0.2.5 release (#189) (@ablaom)
• Combined all layers into hidden and changed hyperparameters to match sklearn defaults (#190) (@john-waczak)

v0.2.4

MLJFlux v0.2.4

Diff since v0.2.3

Closed issues:

• Showcase integration with MLJIteration.jl (#103)
• MLJFlux super slow / stackoverflow (#185)

Merged pull requests:

• Improve one-hot encoding performance (#186) (@ablaom)
• For a 0.2.4 release (#187) (@ablaom)

v0.2.3

MLJFlux v0.2.3

Diff since v0.2.2

Closed issues:

• Regularisation parameters alpha and lambda are not used (#169)
• Re-instante depwarnings in CI (#182)

Merged pull requests:

• Add L2/L1 regularization (#179) (@ablaom)
• Disable depwarns in ci.yml (#183) (@ablaom)
• For a 0.2.3 release (take II) (#184) (@ablaom)

v0.2.2

MLJFlux v0.2.2

Diff since v0.2.1

Merged pull requests:

• Evaluate mnist notebook (#174) (@ablaom)
• MNIST Example update. No new release (#175) (@ablaom)
• Fix epochs in mnist example (plots showing half correct value) (#176) (@ablaom)
• Automate creation of new builders with @build (#177) (@jbrea)
• For a 0.2.2 release (#180) (@ablaom)

v0.2.1

MLJFlux v0.2.1

Diff since v0.2.0

Closed issues:

• Add rng to model hyperparameters and to builder signature (#160)

Merged pull requests:

• Update mnist example (#171) (@ablaom)
• Address ScientificTypes renaming (#172) (@ablaom)
• For a 0.2.1 release (#173) (@ablaom)

v0.2.0

MLJFlux v0.2.0

Diff since v0.1.17

• (enhancement) Add rng hyper-parameter to all models, for passing to builders (#166)
• (breaking) Extend the signature of build to include an RNG, as in build(builder, rng, shape...) (#166)

Merged pull requests:

• Refactor fit and update (#161) (@ablaom)
• Add interface point for RNGs (#166) (@ablaom)
• For a 0.2 release (#168) (@ablaom)
• Fix faulty test leading to "scalar indexing issue" on GPU builds (#170) (@ablaom)

v0.1.17

MLJFlux v0.1.17

Diff since v0.1.16

Merged pull requests:

• Correclty load MultitargetNeuralNetworkRegressor (#158) (@Leonardbcm)

v0.1.16

MLJFlux v0.1.16

Diff since v0.1.15

Closed issues:

• Fix misleading CI badges (#140)
• Stability of MLJFlux.jl API (#151)
• Silent Bug during loss computation (#152)

Merged pull requests:

• Make sure the nightly CI has a different workflow name (#149) (@DilumAluthge)
• Make sure the nightly CI has a different workflow name (#149) (#150) (@ablaom)
• Redefine reformat for Continuous vectors to address silent loss function bug (#153) (@ablaom)
• Fix predict functions of models using BatchNorm (#154) (@Leonardbcm)
• add comment (#156) (@ablaom)

v0.1.15

MLJFlux v0.1.15

Diff since v0.1.14

Merged pull requests:

• Bump compat ProgressMeter = "1.7.1" and version = "0.1.15" (#147) (@ablaom)