|
2 | 2 |
|
3 | 3 | ## Unreleased |
4 | 4 |
|
| 5 | +### Added |
| 6 | + |
| 7 | +- Support for `torch.Tensor` as underlying data type in `Dataset` and |
| 8 | + `GroupedDataset` |
| 9 | + [PR #673](https://github.com/aai-institute/pyDVL/pull/673) |
| 10 | +- Support for pytorch models in most valuation methods when wrapped in |
| 11 | + classes implementing the protocol `TorchSupervisedModel`, e.g. by using |
| 12 | + [skorch.NeuralNetClassifier](https://skorch.readthedocs.io/en/stable/classifier.html) |
| 13 | + models |
| 14 | + [PR #673](https://github.com/aai-institute/pyDVL/pull/673) |
| 15 | +- Improved documentation and examples using DeepSets for Data Utility Learning |
| 16 | + [PR #674](https://github.com/aai-institute/pyDVL/pull/674) |
| 17 | + |
| 18 | +### Fixed |
| 19 | + |
| 20 | +- Issues with `Dataset` indexing |
| 21 | + [PR #673](https://github.com/aai-institute/pyDVL/pull/673) |
| 22 | + |
| 23 | +### Changed |
| 24 | + |
| 25 | + |
| 26 | +## v0.10.0 - 💥📚🐞🆕 New valuation interface, improved docs, new methods, breaking changes and tons of improvements |
| 27 | + |
5 | 28 |
|
6 | 29 | ### Added |
7 | 30 |
|
| 31 | +- Simple result serialization to resume computation of values |
| 32 | + [PR #666](https://github.com/aai-institute/pyDVL/pull/666) |
8 | 33 | - Simple memory monitor / reporting |
9 | 34 | [PR #663](https://github.com/aai-institute/pyDVL/pull/663) |
10 | 35 | - New stopping criterion `MaxSamples` |
11 | 36 | [PR #661](https://github.com/aai-institute/pyDVL/pull/661) |
| 37 | +- New pseudo-stopping criterion `History` to record value updates during fitting |
| 38 | + [PR #653](https://github.com/aai-institute/pyDVL/pull/653) |
12 | 39 | - Introduced `UtilityModel` and two implementations `IndicatorUtilityModel` |
13 | 40 | and `DeepSetsUtilityModel` for data utility learning |
14 | 41 | [PR #650](https://github.com/aai-institute/pyDVL/pull/650) |
|
29 | 56 | - Refactor MSR Banzhaf semivalues with the new sampler architecture. |
30 | 57 | [PR #605](https://github.com/aai-institute/pyDVL/pull/605) |
31 | 58 | [PR #641](https://github.com/aai-institute/pyDVL/pull/641) |
| 59 | + [PR #653](https://github.com/aai-institute/pyDVL/pull/653) |
32 | 60 | - Refactor group-testing shapley values with new sampler architecture |
33 | 61 | [PR #602](https://github.com/aai-institute/pyDVL/pull/602) |
34 | 62 | - Refactor least-core data valuation methods with more supported sampling |
|
89 | 117 |
|
90 | 118 | ### Changed |
91 | 119 |
|
| 120 | +- Major rewrite of the documentation for the new interface in 0.10.0 |
| 121 | + [PR #653](https://github.com/aai-institute/pyDVL/pull/653) |
92 | 122 | - Slicing, comparing and setting of `ValuationResult` behave in a more |
93 | | - natural way |
94 | | - [PR #660](https://github.com/aai-institute/pyDVL/pull/660) |
| 123 | + natural and consistent way. Several bug fixes. |
| 124 | + [PR #653](https://github.com/aai-institute/pyDVL/pull/653) |
| 125 | + [PR #660](https://github.com/aai-institute/pyDVL/pull/660) |
| 126 | + [PR #666](https://github.com/aai-institute/pyDVL/pull/666) |
95 | 127 | - Switched all semi-value coefficients and sampler weights to log-space in |
96 | 128 | order to avoid overflows |
97 | 129 | [PR #643](https://github.com/aai-institute/pyDVL/pull/643) |
|
160 | 192 | - renaming `PreConditioner` -> `Preconditioner` |
161 | 193 | - fit to `TensorOperator` |
162 | 194 | [PR #601](https://github.com/aai-institute/pyDVL/pull/601) |
| 195 | + - Bumped `zarr` dependency to v3 |
| 196 | + [PR #668](https://github.com/aai-institute/pyDVL/pull/668) |
163 | 197 |
|
164 | 198 |
|
165 | 199 | ## 0.9.2 - 🏗 Bug fixes, logging improvement |
|
0 commit comments