This PR does not push any meaningful implementation, just create some modules in the project.
dattri.func: This module contains the modularized low-level implementation of dattri implementations. ihvp and random_projection are our current modularized parts.
dattri.model_utils: For model invasion or other helper function on model.
dattri.algorithm: This module contains the Attributor API for different families of data attribution methods.
dattri.datasets: This module contains the benchmark dataset and tasks (pre-prepared by us). The module name and function design will follow sklearn.datasets.
dattri.metrics: This module contains the evaluation function (e.g., loo(score, target), lds(score, target), ...). The score should be the attribution output generated by dattri.algorithm and the target should be the exact ground-truth provided by dattri.datasets, users, or some helper function we provide.
This PR does not push any meaningful implementation, just create some modules in the project.
dattri.func: This module contains the modularized low-level implementation of dattri implementations.ihvpandrandom_projectionare our current modularized parts.dattri.model_utils: For model invasion or other helper function on model.dattri.algorithm: This module contains theAttributorAPI for different families of data attribution methods.dattri.datasets: This module contains the benchmark dataset and tasks (pre-prepared by us). The module name and function design will followsklearn.datasets.dattri.metrics: This module contains the evaluation function (e.g.,loo(score, target),lds(score, target), ...). Thescoreshould be the attribution output generated bydattri.algorithmand thetargetshould be the exact ground-truth provided bydattri.datasets, users, or some helper function we provide.