MLJFlux.jl allows users to quickly build, train and optimise Flux.jl neural network models using the high-level, general purpose machine learning framework MLJ.jl.
The new 0.5 release assimilates a number of substantial under-the-hood improvements, such as:
- better performing re-implementation of L1-L2 regularization
- implicit-style AD
- use of Optimiser.jl optimisers
It also provides a new model, NeuralNetworkBinaryClassification
, an optimised version of the existing classifier for binary targets.
MLJFlux.jl 0.5 sports substantially revamped documentation including new sample workflows and extended examples.
Thanks to Tiem van der Deure, @EssamWisam, @pat-alt for contributions to this release.