CHANGELOG for 0.12.0 release (pytorch#2522)

Summary:
Pull Request resolved: pytorch#2522

changelog

Differential Revision: D62405336

CHANGELOG.md

@@ -2,6 +2,60 @@
 
 The release log for BoTorch.
 
+## [0.12.0] -- Sep 9, 2024
+
+#### Major changes
+* Update default hyperparameter priors for most models to use dimension-scaled log-normal
+  priors, which make performance much more robust to dimensionality. See
+  discussion #2451 for details. The only models that are not changed are fully
+  Bayesian models, `PairwiseGP`, and fidelity kernels for multi-fidelity models;
+  for models that utilize a composite kernel, such as
+  multi-fidelity/task/context, this change only affects the base kernel (#2449,
+  #2450, #2507).
+* Use `Standarize` by default with all the models using the upgraded priors. In
+  addition to reducing the amount of boilerplate needed to initialize a model,
+  this change was motivated by the change to default priors, because the new
+  priors will work less well when data is not standardized. Users who do not
+  want to use transforms should explicitly pass in `None` (#2458, #2532).
+
+#### Compatibility
+* Unpin Numpy (#2459).
+* Require PyTorch>=2.0.1, GPyTorch==1.13, and linear_operator==0.5.3 (#2511).
+
+#### New features
+* Introduce `PathwiseThompsonSampling` acquisition function (#2443).
+* Enable `qBayesianActiveLearningByDisagreement` to accept a posterior
+  transform, and improve its implementation (#2457).
+* Enable `SaasPyroModel` to sample via NUTS when training data is empty (#2465).
+* Add multi-objective `qBayesianActiveLearningByDisagreement` (#2475).
+* Add input constructor for `qNegIntegratedPosteriorVariance` (#2477).
+* Introduce `qLowerConfidenceBound` (#2517).
+
+#### Bug fixes
+* Fix `batch_shape` default in `OrthogonalAdditiveKernel`.
+* Ensure all tensors are on CPU in `HitAndRunPolytopeSampler` (#2502).
+* Fix duplicate logging in `generation/gen.py` (#2504).
+* Raise exception if `X_pending` is set on underlying `AcquisitionFunction` in
+  prior-guided `AcquisitionFunction` (#2505).
+* Affine input transforms error with data of incorrect dimension, even in eval
+  mode (#2510).
+
+#### Documentation
+* Clarify incompatibility of inter-point constraints with `get_polytope_samples`
+  (#2469).
+* Update tutorials to use the log variants of EI-family acquisition functions,
+  don't make tutorials pass `Standardize` unnecessarily, and other
+  simplifications and cleanup (#2462, #2463, #2490, #2495, #2496, #2498, #2499).
+
+#### Other changes
+* More informative warnings about failure to standardize or normalize data
+  (#2489).
+* Suppress irrelevant warnings in `qHypervolumeKnowledgeGradient` helpers
+  (#2486).
+* Cleaner `botorch/acquisition/multi_objective` directory structure (#2485).
+* Remove deprecated argument `data_fidelity` to `SingleTaskMultiFidelityGP` and
+  deprecated model `FixedNoiseMultiFidelityGP` (#2532).
+
 ## [0.11.3] -- Jul 22, 2024
 
 #### Compatibility