Miguel de Benito Delgado

Ideas: - Mislabeled data detection - Active learning / subset selection (compare with random)

Branch feature/loo https://github.com/appliedAI-Initiative/pyDVL/compare/develop...feature/loo implements LOO for linear smoothers, for which a closed form solution is known. Finish that and test.

Interruptible samplers

1

#319 introduces `PermutationSampler` but it does not include the possibility of interrupting the sampling within a permutation, as required for TCMS. One possibility would be to make samplers not simple...

Introduced in _Zhaoxuan Wu, Yao Shu, and Bryan Kian Hsiang Low, “[DAVINZ: Data Valuation Using Deep Neural Networks at Initialization](https://proceedings.mlr.press/v162/wu22j.html),” in Proceedings of the 39th International Conference on Machine Learning...

* [ ] Implement the optimal strategy described in [1]. * [x] #226 * [ ] Reproduce their results (possibly, but not necessarily including the stratified sampling strategy of [2]),...

tbd See - [ ] #325 - [ ] #463

#319 implements Banzhaf indices and Beta Shapley, but only superficially documents them. We need a notebook comparing them to other methods, along the lines of the experiments in the respective...

- [ ] #472 - [ ] #469 - [ ] #471 - (see #234) - [ ] Finish #341 - [ ] #470

Use any approach that allows the caller to compose the plots together without side-effects. This should be done without requiring calling code to add boilerplate. Typically one passes the axes...