Now works with keras > 2.3 and tensorflow > 2.3.
Generators
- Re-worked generators, now use a python implementation. This required re-designing generator constructors and so on.
Learners
- Shaped MLP 1 & 2 Learners for Regression and Classification
- mlr3keras can now deal with images via the new
KerasCNN learner.
- Added a deep-wide architecture inspired by Ericson et al., 2020 AutoGluon-Tabular: Robust and Accurate AutoML for Structured Data.
Bugfixes
- TabNet now defaults to
num_layers = 1L for stack = TRUE.
Learners
- TabNet and FeedForward can now deal with factor / ordered / character features
- FeedForward Keras Models now default to “embeddings” for factor features
General
- KerasArchitecture: Introduced new abstraction for architectures. This should rarely be visible to users but makes stuff easier to extend.
Learners
- Add regression learner for a custom model
- Add regression learner for parametrized feedforward model
- Add stacked and unstacked tabnet classification and regression learner
