tabnet 0.8.0

New features

messaging is now improved with {cli}
add optimal threshold and support size into new 1.5 alpha entmax15() and sparsemax15() mask_types. Add an optional mask_topk config parameter. (#180)
optimizernow default to the torch_ignite_adam when available. Result is 30% faster pretraining and fitting tasks (#178).
add nn_aum_loss() function for area under the \(Min(FPR,FNR)\) optimization for cases of unbalanced binary classification (#178).
add a vignette on imbalanced binary classification with nn_aum_loss() (#178).

config parameter now merge correctly for torch loss or torch optimizer generator.
nn_unsupervised_loss() is now a proper loss function.

Remove long-run example raising a Note.
fix tabet_pretrain failing with value_error("Can't convert data of class: 'NULL'") in R 4.5
fix tabet_pretrain wrongly used instead of tabnet_fit in Missing data predictor vignette
improve message related to case_weights not being used as predictors.
improve function documentation consistency before translation.
fix “…” is not an exported object from ‘namespace:dials’” error when using tune() on tabnet parameters. (#160 @cphaarmeyer)

parsnip models now allow transparently passing case weights through workflows::add_case_weights() parameters (#151)
parsnip models now support tabnet_model and from_epoch parameters (#143)

Adapt tune::finalize_workflow() test to {parsnip} v1.2 breaking change. (#155)
autoplot() now position the “has_checkpoint” points correctly when a tabnet_fit() is continuing a previous training using tabnet_model =. (#150)
Explicitely warn that tabnet_model option will not be used in tabnet_pretrain() tasks. (#150)

{tabnet} now allows hierarchical multi-label classification through {data.tree} hierarchical Node dataset. (#126)
tabnet_pretrain() now allows different GLU blocks in GLU layers in encoder and in decoder through the config() parameters num_idependant_decoder and num_shared_decoder (#129)
Add reduce_on_plateau as option for lr_scheduler at tabnet_config() (@SvenVw, #120)
use zeallot internally with %<-% for code readability (#133)
add FR translation (#131)

Add explicit legend in autoplot.tabnet_fit() (#67)
Improve unsupervised vignette content. (#67)
tabnet_pretrain() now allows missing values in predictors. (#68)
tabnet_explain() now works for tabnet_pretrain models. (#68)
Allow missing-values values in predictor for unsupervised training. (#68)
Improve performance of random_obfuscator() torch_nn module. (#68)
Add support for early stopping (#69)
tabnet_fit() and predict() now allow missing values in predictors. (#76)
tabnet_config() now supports a num_workers= parameters to control parallel dataloading (#83)
Add a vignette on missing data (#83)
tabnet_config() now has a flag skip_importance to skip calculating feature importance (@egillax, #91)
Export and document tabnet_nn
Added min_grid.tabnet method for tune (@cphaarmeyer, #107)
Added tabnet_explain() method for parsnip models (@cphaarmeyer, #108)
tabnet_fit() and predict() now allow multi-outcome, all numeric or all factors but not mixed. (#118)

tabnet_explain() is now correctly handling missing values in predictors. (#77)
dataloader can now use num_workers>0 (#83)
new default values for batch_size and virtual_batch_size improves performance on mid-range devices.
add default engine="torch" to tabnet parsnip model (#114)
fix autoplot() warnings turned into errors with {ggplot2} v3.4 (#113)

Added an update method for tabnet models to allow the correct usage of finalize_workflow (#60).

Allow model fine-tuning through passing a pre-trained model to tabnet_fit() (@cregouby, #26)
Explicit error in case of missing values (@cregouby, #24)
Better handling of larger datasets when running tabnet_explain().
Add tabnet_pretrain() for unsupervised pretraining (@cregouby, #29)
Add autoplot() of model loss among epochs (@cregouby, #36)
Added a config argument to fit() / pretrain() so one can pass a pre-made config list. (#42)
In tabnet_config(), new mask_type option with entmax additional to default sparsemax (@cmcmaster1, #48)
In tabnet_config(), loss now also takes function (@cregouby, #55)