dattri release 0.2.0

What's Changed

• [dattri.metrics] Add brittleness test by @SeanZh30 in #114
• [dattri] backward compatibility by @TheaperDeng in #125
• [dattri.algorithm] Add KNN data shapley by @sx-liu in #122
• [dattri.algorithm] Fix misalignment in KNN shapley by @sx-liu in #126
• [dattri.task] add partial parameter support for target/loss function and their gradient func by @TheaperDeng in #118
• Improve docstring in dattri.func and dattri.metrics by @jiaqima in #127
• [dattri.metrics] change the shape of score in lds to improve consistency by @TheaperDeng in #132
• [dattri.algorithm] Refactoring of BaseInnerProductAttributor by @TheaperDeng in #133
• Fix typos and improve docstrings by @jiaqima in #134
• [dattri.func] Re-implementation of Arnoldi projector by @tingwl0122 in #135
• [dattri experiments] benchmark script for mnist, cifar and shakespeare by @TheaperDeng in #128
• [dattri.algorithm] add partial parameters support for IF, TracIN, Grad-Dot/Cos, TRAK by @TheaperDeng in #137
• Rename the modules metric, model_util, and script by @jiaqima in #139
• [dattri.task] Rename index in task to ckpt_idx by @jiaqima in #141
• [dattri.algorithm] relax the requirement to the loss_func. by @TheaperDeng in #140
• [Examples] refactor examples and add github action to automatically validate the scripts. by @TheaperDeng in #138
• [dattri.algorithm] Changes to the implementation of DataInf. by @charles-pyj in #123
• [dattri.experiments] Add MusicTransformer experiment by @tingwl0122 in #142
• Add paper info by @jiaqima in #144
• Update citation format by @sleepymalc in #145
• [Document] support multiple document at the same time. by @TheaperDeng in #147
• [installation] fix installation and some name typo by @TheaperDeng in #146
• [dattri.algorithm, dattri.func] Refactor the implementation of EKFAC by @sx-liu in #143
• Minor improvements on READMEs and docs by @jiaqima in #149
• [branch-0.2 readme] fix installation method in readme by @TheaperDeng in #151
• [branch-0.2] Update the version number by @TheaperDeng in #152

New Contributors

• @sleepymalc made their first contribution in #145

Full Changelog: v0.1.0...v0.2.0

dattri release 0.1.0

What's Changed

• create folders by @jiaqima in #1
• Push project architect by @TheaperDeng in #2
• Change some architecture design by @TheaperDeng in #3
• CI/CD setup and direct ihvp function by @TheaperDeng in #4
• switch {pycodestlye, pydocstyle, pylint} to ruff by @jiaqima in #7
• add darglint and refactor ihvp.py to follow google style docstring by @jiaqima in #8
• [dattri.func] add hvp and ihvp_cg by @TheaperDeng in #5
• Benchmark API signature (w/o PBRF) by @TheaperDeng in #13
• Initiate projector code and test file by @tingwl0122 in #12
• Add sphinx docs by @xingjian-zhang in #19
• Implement flip_label and mislabel_detection_auc by @sx-liu in #14
• Implemented retrain_loo and calculate_loo_groundtruth. by @charles-pyj in #16
• Enable ruff opt-in preview mode. by @xingjian-zhang in #23
• Add a new pull request template to repo by @TheaperDeng in #6
• Model training functions and loss calculation functions for mnist + lr by @TheaperDeng in #25
• Support target function which contains torch.func.functional_call by @TheaperDeng in #18
• Allow additional package (fast_jl) installation in setup.py by @tingwl0122 in #27
• Change ruff check rules by @jiaqima in #34
• Fix PR template by @jiaqima in #36
• Add pre-commit and CONTRIBUTING.md by @jiaqima in #37
• Migrate setup to pyproject by @jiaqima in #38
• Add makefile by @jiaqima in #39
• clean up runners by @jiaqima in #41
• remove darglint by @jiaqima in #42
• fix some missing parts in mnist_lr retrain func by @TheaperDeng in #44
• relax the test tol for test_ihvp_cg_nn by @TheaperDeng in #45
• Re-add darglint by @jiaqima in #46
• minor change in mnist and fix test file positions by @jiaqima in #47
• [dattri.model_utils] implement retrain_lds by @jiaqima in #48
• [dattri.func.random_projection] Add a wrapper to decide projector class by @tingwl0122 in #20
• [dattri.metrics] Implement LDS by @jiaqima in #49
• [dattri.func.ihvp] Add Arnoldi as another ihvp method by @tingwl0122 in #33
• [dattri.func] update cg ihvp to support vmap by @TheaperDeng in #61
• [dattri.benchmark] Add MAESTRO benchmark by @tingwl0122 in #53
• [dattri.benchmark] refactor the benchmark code for mnist by @TheaperDeng in #56
• [dattri.func] Fixing Arnoldi's for-loop implementation by @tingwl0122 in #59
• [dattri.benchmark] Add a dataset folder by @tingwl0122 in #64
• [dattri.model_utils] Add a wrapper for register_forward_hook by @tingwl0122 in #50
• [dattri.benchmark] upload the imangenet training script as well as a dattri_retrain command by @TheaperDeng in #51
• [dattri.benchmark] Modify MAESTRO exp to fit current structure by @tingwl0122 in #67
• [dattri.benchmark] Add nanoGPT and retrain function here by @SeanZh30 in #60
• [dattri.algorithm] IF attributor for exisiting ihvp functions by @TheaperDeng in #32
• [dattri.algorithm] Add TracIn Attributor by @tingwl0122 in #54
• [dattri.func.random_projection] Format/naming/bugs fixing in related files by @tingwl0122 in #63
• [dattri.benchmark] Add a new benchmark model MLP by @TheaperDeng in #70
• [dattri.benchmark] Add ResNet-9 benchmark on CIFAR-2 by @tingwl0122 in #68
• [dattri.benchmark] Fix output format of maestro dataset creation by @tingwl0122 in #69
• [dattri.algorithms] add arnoldi support in IFAttributor by @TheaperDeng in #74
• [dattri.func] Implement ihvp_lissa and ihvp_at_x_lissa by @sx-liu in #30
• [dattri.benchmark] add a subsetsampler for easy usage by @TheaperDeng in #66
• [dattri.algorithm] Modify TracIn score computation loop by @tingwl0122 in #72
• [dattri.benchmark] Add CIFAR-2 and MAESTRO support in dattri/scripts/dattri_retrain.py by @tingwl0122 in #77
• [dattri.model_utils] Add utils for dropout by @tingwl0122 in #65
• [dattri] fix pyproject.toml by @TheaperDeng in #84
• [dattri.benchmark] Add TracIn example (noisy label on MNIST and CIFAR-2) by @tingwl0122 in #78
• [dattri.algorithm] Adapt vmap to IFAttributor by @tingwl0122 in #81
• [dattri.algorithm] Influence Function LiSSA option by @TheaperDeng in #80
• [dattri.algorithm] Add TRAKAttributor and related examples by @TheaperDeng in #83
• [dattri.func.ihvp] Implementation of DataInf by @charles-pyj in #82
• [dattri.algorithm] Fix trak problems by @TheaperDeng in #92
• Add Readme by @TheaperDeng in #93
• [dattri.func.ihvp] Implement ihvp_at_x_ekfac by @sx-liu in #94
• Add write permissions for doc workflow. by @xingjian-zhang in #95
• [Doc] add api docs by @TheaperDeng in #96
• [dattri.benchmark] Correcting the performance of (cifar2) benchmark experiment by @tingwl0122 in #97
• [dattri.algorithm] Re-add RPS attributor by @tingwl0122 in #98
• [dattri.algorithm] Add AttributionTask abstraction for new API design by @TheaperDeng in #99
• [Document] update the installation info by @TheaperDeng in #101
• [dattri.matrics] add loo_corr by @TheaperDeng in #102
• [dattri.func] rename ihvp.py -> hessian.py/fisher.py by @TheaperDeng in #103
• [dattri] Move some files to better path by @TheaperDeng in #105
• [dattri.algorithm] Support AttributionTask API in TracIN/TRAK/RPS Attributors by @TheaperDeng in #104
• [dattri.func] Re-define Arnoldi algorithm as a projection method by @tingwl0122 in #108
• [dattri.benchmark] add benchmark downloading API by @TheaperDeng in #106
• [dattri.task] Add loss_func and make target_func optional by @TheaperDeng in #111
• [dattri.algorithm] Fix Arnoldi algorithm's bug by @tingwl0122 in #112
• Add Logo by @TheaperDeng in #115
• [dattri.func] Fix ChunkedCudaProjector bug by @tingwl0122 in #113
• [Doc] Quick fix on the document about __init__ methods' correct presentation. by @TheaperDeng in #116
• [branch-0.1] fix pyproject.toml by @TheaperDeng in #117

New Contributors

• @jiaqima made their first contribution in #1
• @TheaperDeng made their first contribution in #2
• @tingwl0122 made their first contribution in #12
• @xingjian-zhang made their first contribution in #19
• @sx-liu made their first contribution in #14
• @charles-pyj made their first contribution in #16
• @SeanZh30 made their first contribution in #60

Full Changelog: https://github.com/TRAIS-Lab/dattri/commits/v0.1.0