Releases: ENSTA-U2IS-AI/torch-uncertainty
Releases · ENSTA-U2IS-AI/torch-uncertainty
v0.7.0 Time Series Classification Support
What's Changed
- ✨ Time-Series Classification support by @alafage & @o-laurent in #194
- Add InceptionTime model (single, mimo, packed-ensembles, bnn, etc.)
- Add UCRUEADataset and UCRUEADataModule
- Add Experiment config files
- ✨ Add complexity metrics (number of parameters and floating-point operations) in all routines by @alafage in #199
- 🔨 Change default implementation value in
PackedLinear
to"einsum"
as it seems faster.
Maintenance updates
- 🔥 Remove PyPI token by @o-laurent in #195
- 🔧 Check that torch-uncertainty can be loaded without optional dependencies by @o-laurent in #198
- ⚡ Update ruff by @o-laurent in #200
- 🔧 Update pyproject by @o-laurent in #202
- 🔒 Restrict releases by @o-laurent in #205
Full Changelog: v0.6.0...v0.7.0
v0.7.0.post1 Make tslearn truly optional
What's Changed
- 🐛 Fix unprotected optional dependency by @o-laurent in #210
Full Changelog: v0.7.0...v.0.7.0.post0
v0.6.0 Regression update
What's Changed
- 🔨 Update Regression support by @alafage in #180
- ✨ Add QuantileCalibrationError metric
- 🔨 MC Dropout supports probabilistic regression
- 🔨 Stochastic Model (BNN) supports probabilistic regression
- ✨ Add more MLP model versions (e.g., MIMO, BatchEnsemble, etc.)
Full Changelog: v0.5.2...v0.6.0
v0.5.2.post0 PackedLayerNorm fix
v0.5.2 Rework Frost data
What's Changed
- 🔨 Rework Frost as external dependency by @o-laurent in #183
- 🔨 Rework FrostImages dataset & Rename a plotting function by @o-laurent in #184
Full Changelog: v0.5.1...v0.5.2
v0.5.1 :hammer: Rework Conformal methods and fix scalers in the multiclass setting
Pre-release
What's Changed
- ⚡ Version 0.5.1 by @o-laurent in #178
- Fixed an incorrect behavior in the multiclass setting for TemperatureScalers
- Reworked Conformal prediction methods and acknowledged TorchCP
- Also improved the tutorials to some extent
Full Changelog: v0.5.0...v0.5.1
v0.5.0 Major Update
What's Changed
- 📚 New documentation by @fira7s in #128
- 🐛 Fix erronous experimental config for Bayesian LeNet by @tonyzamyatin in #132
- ⬆️ Bump tj-actions/changed-files from 42 to 46 in /.github/workflows by @dependabot in #148
- 🔧 Run docs on self-hosted by @o-laurent in #149
- 🐛 Fix for Lightning 2.5 by @alafage in #150
- 🔧 Try fixing SSH agent issue by @o-laurent in #151
- ✨ Add an evalaluation batch size argument by @o-laurent in #156
- 🔨 Rework OOD criteria by @o-laurent in #153
- 📚 Update Dockerfile documentation by @tonyzamyatin in #161
- ✨ Add routine specific checkpoint callbacks for uncertainty-aware training by @alafage in #166
- ✨ Add Conformal prediction support by @giannifranchi in #168
- ✨ Add Test-Time Adaptation by @o-laurent in #154
- ✨ Add SGMCMC optimizers and sampling methods by @alafage in #170
- ✨ Enable combining val metrics by @alafage in #169
- ✨ Enable preprocessing without model by @o-laurent in #172
- 🎨 Revamp Tutorials by @alafage in #155
- ✨ Everything as a second level import by @o-laurent in #173
- 🐛 Support binary scaling by @o-laurent in #174
- 📚 Revamp the documentation by @o-laurent in #142
- 🔨 SegmentationRoutine supports OOD detection by @alafage in #177
- ✨ Version 0.5.0 by @o-laurent in #160
Full Changelog: v0.4.3...v0.5.0
v0.4.3.rc0 Compatibility fix for Lightning 2.5.1
What's Changed
- ⬆️ Bump tj-actions/changed-files from 42 to 46 in /.github/workflows by @dependabot in #148
- 🐛 Fix for Lightning 2.5.1 by @alafage in #150
Full Changelog: v0.4.3...v0.4.3.post0
Warning
Lightning<2.5.1 not supported for the CLI, if you encounter any issue please reach out to us.
v0.4.3 Rework PostProcessings, Enable dataset corruptions, use transforms v2
Pre-release
What's Changed
- 🔨 Rework post-processing classes to take a
torch.utils.data.DataLoader
in their.fit()
methods by @o-laurent, @alafage, @fira7s - 🔨 Improve the corruption transforms by @o-laurent in #144
- 🔨 All transforms rely on
torchvision.transforms.v2
by @o-laurent - 🔧 User changed-files again by @o-laurent in #147
Full Changelog: v0.4.2...v0.4.3
v0.4.2 Add docker, Rework BatchEnsemble & misc
What's Changed
- 🐛 Fix MNIST test dataloader for shifted data by @tonyzamyatin in #135
- 🐳 Add Dockerfile and documentation for using Docker image by @tonyzamyatin in #134
- 🔨 Use
kornia
's motion blur instead ofWand
's Gaussian motion blur by @o-laurent in #140 - ✨ LeNet BatchEnsemble and Deep Ensemble + minor bugfixes by @tonyzamyatin in #137
- ✨ Docker settings, BatchEnsemble utilities rework + minor fixes by @alafage in #141
New contributor
Thank you, @tonyzamyatin, for your help adding features and finding small bugs!
Full Changelog: v0.4.1...v0.4.2