Optional mlr3 learner wrappers for modern tabular methods commonly used as strong baselines on biomarker-style classification problems. The wrappers are designed for small-\(P\), medium-\(N\), and class-imbalanced omics tasks.
Details
The learners in this file are intentionally lazy about backend availability: they can be constructed without the Python or CatBoost runtime being present, but training will fail with an actionable error if the required dependency is missing.
Backends: - `make_tabnet_learner()`: `pytorch_tabnet` through `reticulate` - `make_tabm_learner()`: `tabm` or `rtdl_revisiting_models` through `reticulate` - `make_fttransformer_learner()`: `rtdl_revisiting_models` through `reticulate` - `make_catboost_learner()`: `catboost` R package - `make_tabpfn_learner()`: `tabpfn` through `reticulate`
GPU selection follows `torch.cuda.is_available()` when `device = "auto"`.