search

scikit-adaptation / skada

Domain adaptation toolbox compatible with scikit-learn and pytorch
https://scikit-adaptation.github.io/
BSD 3-Clause "New" or "Revised" License
54 stars 16 forks source link

Propogate `sample_weight` for resampling adapters through multiple steps #146

Closed kachayev closed 3 months ago

kachayev commented 4 months ago

With the current implementation of how AdaptationOutput is propagated through the pipeline we should have troubles when reweighting adapter (producer of weights) and the estimator (consumer) are separated, e.g. when there's another transformer in between of them.