Import fix

Fixes some imports on a clean install.

v0.1.0 — Initial alpha release

incerto v0.1.0 — first public alpha of a comprehensive uncertainty quantification library for PyTorch.

Warning

Alpha release. The public API may change without notice before v1.0.

What's included

• Calibration — Temperature/Vector/Matrix/Dirichlet scaling, Platt, Isotonic,
  Histogram binning, Beta calibration; Label smoothing, Focal Loss,
  Confidence penalty, Evidential DL; 7 metrics including smECE.
• OOD detection — MSP, MaxLogit, Energy, ODIN, Mahalanobis, KNN; Mixup,
  CutMix, Outlier Exposure, Energy regularization.
• Conformal prediction — ICP, APS, RAPS, Mondrian (classification);
  Jackknife+, CV+, CQR (regression).
• Selective prediction — Softmax threshold, Deep Gambler, SelectiveNet, SAT.
• Bayesian methods — MC Dropout, Deep Ensembles, SWAG, Laplace, Variational
  BNN; uncertainty decomposition.
• Active learning — 6 acquisition functions, 5 query strategies (incl. BADGE,
  Core-Set, BatchBALD).
• Distribution shift — MMD, Energy, KS, Classifier (BBSD); label shift +
  importance weighting.
• LLM uncertainty — Token entropy/confidence/perplexity; sequence-level
  probability/entropy; sampling (self-consistency, semantic entropy, MI);
  verbalized (PTrue, self-evaluation, bidirectional consistency).

Tested with

• Python 3.10, 3.11, 3.12, 3.13
• PyTorch ≥ 2.0, NumPy ≥ 1.24, scikit-learn ≥ 1.3, scipy ≥ 1.11
• 982 tests passing across Linux and macOS

Install

pip install incerto                  # core
pip install incerto[vision]          # + torchvision for vision benchmarks
pip install incerto[llm]             # + transformers, accelerate
pip install incerto[all]             # everything

See the examples/ directory for 8 Jupyter tutorials covering every module end-to-end.

Full changelog: CHANGELOG.md