## [0.2.0] - 2026-03-28
### Fixed
- Parameter handling got more robust, with better error messages for invalid inputs and edge cases.
- Replaced scale-dependent absolute slope threshold with Pearson correlation > 0.5 combined with a peak cross-map skill > 0.3, matching the Sugihara et al. (2012) convergence criterion.

### Changed
- Switched from inverse-distance weighting to canonical exponential weights $w_i = \exp(-d_i / d_{\min})$ as specified in Sugihara et al. (2012).

### Performance
- Run both causal directions concurrently.

## [0.1.3] - 2025-10-21
### Fixed
- Validate inputs in `CCM.new/3` (embedding_dim, tau, num_samples) and clamp computed library size for short series.
- Make weighting scheme deterministic and numerically stable: use inverse-distance normalized weights; exact matches dominate.

### Added
- Test-only wrapper `CCM.predict_point_for_tests/3` to allow unit testing of internal prediction logic.
- New unit tests covering edge cases and prediction weighting behavior.

## [0.1.2] - 2025-08-24
### Changed
- Minor performance optimisation
- Changed the format of math expressions in README.md

## [0.1.1] - 2025-08-21
### Changed
- Minor refactor

## [0.1.0] - 2025-08-07
### Added
- Initial release