Skip to content

CHANGELOG.md

All notable changes to uncertainty_flow are documented in the root CHANGELOG.md.

This file is kept for historical reference. The canonical changelog is at the repository root.

[Unreleased]

Added

  • Initial project scaffold and package structure
  • BaseUncertaintyModel abstract base class
  • DistributionPrediction core output object with .quantile(), .interval(), .mean(), .plot()
  • ConformalRegressor — tabular conformal wrapper for any sklearn estimator
  • ConformalForecaster — temporal-aware conformal wrapper with multivariate support
  • QuantileForestForecaster — quantile regression forest with leaf distribution storage
  • DeepQuantileNet — multi-quantile MLP with shared trunk (sklearn backend)
  • DeepQuantileNetTorch — PyTorch-backed multi-quantile network with GPU and monotonicity loss (optional)
  • TransformerForecaster — Chronos-2 pretrained forecasting wrapper (optional)
  • Polars I/O layer with LazyFrame support (utils/polars_bridge.py)
  • Holdout and cross-conformal calibration split strategies
  • Residual correlation analysis for uncertainty driver detection (uncertainty_drivers_)
  • Gaussian, Clayton, Gumbel, Frank copula families with auto-selection
  • Calibration report (Polars DataFrame) with pinball loss, Winkler score, coverage
  • Warning system with UF-W / UF-E codes
  • Standalone metrics: pinball_loss, winkler_score, coverage_score
  • Shared model persistence via .save() / .load() with .uf archives and portable metadata
  • BayesianQuantileRegressor — NumPyro MCMC with horseshoe priors (optional)
  • CausalUncertaintyEstimator — doubly-robust, S-learner, T-learner treatment effect estimation
  • CrossModalAggregator — per-feature-group models with product/copula/independent aggregation
  • ConformalRiskControl — conformal calibration for user-defined risk functions
  • Built-in risk functions: asymmetric_loss, threshold_penalty, inventory_cost, financial_var
  • UncertaintyExplainer — counterfactual explanations for interval width reduction
  • EnsembleDecomposition — bootstrap-based aleatoric/epistemic uncertainty decomposition
  • FeatureLeverageAnalyzer — per-feature uncertainty attribution with aleatoric/epistemic scores
  • launch_dashboard() — interactive Streamlit calibration dashboard (optional)
  • CLI benchmarking with 108+ HuggingFace datasets, auto-tuning, and structured output
  • auto_tune parameter on supported models for automatic hyperparameter search

Release Notes Template

## [X.Y.Z] — YYYY-MM-DD

### Added
- New features

### Changed
- Changes to existing functionality (non-breaking)

### Deprecated
- Features that will be removed in a future version

### Removed
- Features removed in this release

### Fixed
- Bug fixes

### Security
- Security-relevant changes