Changelog
=========

All notable changes to GRU Dynamic Beta will be documented here.

[0.2.0] - Unreleased
---------------------

Added
~~~~~

* High-level convenience API: ``estimate_beta()``, ``compare_betas()``, ``quick_report()``
* Named configuration presets: default, responsive, smooth, research
* Command-line interface: ``python -m grubeta AAPL SPY``
* Custom exception classes with finance-friendly error messages
* 4 Jupyter notebook tutorials (quickstart, hedging, comparison, research)
* ``format_summary()`` for human-readable beta summaries
* ``grubeta[data]`` optional dependency group for yfinance

Changed
~~~~~~~

* README rewritten with finance-first framing
* Project description updated in pyproject.toml
* Research preset description clarified (enhanced model capacity, not full features via convenience API)
* Walk-forward train window now derives from config instead of hardcoded 500
* TemporalCertificate uses actual package version

[0.1.3] - 2026-02-20
---------------------

Added
~~~~~

* ``lambda_alpha_smooth`` hyperparameter for L2 temporal smoothness penalty on alpha
* 4-component composite loss function (accuracy, beta stability, alpha sparsity, alpha stability)

Changed
~~~~~~~

* ``CAPMLoss.create_composite_loss`` now accepts ``lambda_alpha_smooth`` parameter
* Default ``lambda_alpha_smooth=0.1`` (2x beta smoothness weight, reflecting that alpha should be smoother than beta under CAPM)

[0.1.2] - 2026-02-11
---------------------

Fixed
~~~~~

* Critical: ``pydantic`` dependency now correctly included in published package
* Publish workflow no longer depends on deleted ``testpypi`` environment
* Migrated to Pydantic V2 validators (``@field_validator``, ``ConfigDict``)
* Removed duplicate line in ``_create_return_features``
* Aligned minimum Python version to 3.9 across all metadata
* Added ``initial_beta`` (default=1.0) and ``initial_alpha`` (default=0.0) config parameters
* Removed stray ``verify_revert.py`` from repository

[0.1.1] - 2026-02-06
---------------------

Added
~~~~~

* Fixed missing ``pydantic`` dependency
* Validated "Zero Defects" status
* Improved build configuration

[0.1.0] - 2026-01-23
---------------------

Initial release.

Added
~~~~~

* Core ``DynamicBeta`` class for GRU-based beta estimation
* ``DataPreprocessor`` for feature engineering with lookahead bias prevention
* ``BetaEvaluator`` for comprehensive model evaluation
* Walk-forward validation with anchored expanding window
* Composite loss function (accuracy + stability + sparsity)
* Multiple input modes (simple returns vs. full features)
* Diagnostic tests for lookahead bias detection
* Benchmark comparison utilities (rolling OLS, EWMA, static)
* Model persistence (save/load)
* GPU acceleration support
* Comprehensive documentation and examples

Technical Details
~~~~~~~~~~~~~~~~~

* Dual-pathway GRU architecture (beta + alpha pathways)
* Configurable hyperparameters via ``DynamicBetaConfig``
* Feature engineering via ``FeatureConfig``
* Support for technical indicators (requires ``ta`` library)
* Macroeconomic feature integration

Future Plans
------------

[0.3.0] - Planned
~~~~~~~~~~~~~~~~~

* Attention mechanism for beta pathway
* Transformer-based alternative model
* Regime detection integration
* Real-time prediction mode
* Extended benchmark comparisons

[0.4.0] - Planned
~~~~~~~~~~~~~~~~~

* Multi-asset portfolio beta estimation
* Factor model extension (Fama-French)
* Confidence intervals for beta estimates
* Online learning mode