v0.2.0 (26/02/2026)
Pre-release
Pre-release
Docs-only update: The v0.2.0 Git tag was moved after publication to update the Read the Docs build configuration and documentation index (.readthedocs.yaml, docs/source/index.rst). This does not change the gpyconform==0.2.0 artifacts published on PyPI.
Full Changelog: v0.1.1...v0.2.0
Features
- Inductive (Split) Conformal Prediction (ICP) for GPR: Added ICP support with
GPRICPWrapper(GPyTorch model wrapper) and a model-agnosticInductiveConformalRegressor. - Tensor split utilities: Added
tensor_train_test_splitandtensor_train_cal_splitto conveniently create train/test and proper-train/calibration splits directly ontorch.Tensors.
Updates
- Configurable patching via environment variable: Added
GPYCONFORM_AUTOPATCHto control when/if the internal GPyTorch prediction-strategy patch is applied (eager, lazy, or forbidden). Exposedapply_patches()/is_patched()for manual control.
Fixes
- Precision / numerical stability improvements: Improved numerical robustness in interval construction and confidence-level handling.