state_space 0.1.0

Gaussian process regression for irregular time series: smoothed trends with uncertainty bands, forecasts and learned hyperparameters, in linear time. Pure Dart.

0.1.0

First release.

• Exact Gaussian process regression for irregular time series in linear time: a Kalman filter and RTS smoother over any gaps, repeated timestamps and missing data, reporting the posterior at the observation times or on any output grid.
• Components: LocalLevel, LocalLinearTrend (a natural cubic smoothing spline), TrigonometricSeasonal, RegressionComponent with indicator and step regressors, Matern at ν = 1/2, 3/2 and 5/2, and StochasticCycle.
• fit: maximum marginal likelihood with the noise level concentrated out, a bracket scan before local search, plateau widths per parameter, a measurement-variance floor, warm starts, and warnings in sentences for bounds, flat directions and bad readings.
• StructuralModel.withEstimatedScale for smoothing chosen by the caller, with the noise level estimated from the data.
• forecast, credible and predictive intervals and bands, per-component means and slopes, and regression coefficients with standard errors.
• Exact diffuse initialisation by default, refusing with UnderdeterminedModelException when the data does not determine the model.
• Innovation diagnostics: recursive residuals, autocorrelation and Ljung–Box.
• TimeAxis for converting calendar dates to days and back across changes of clock.
• package:state_space/authoring.dart for writing components, with checkComponent.
• Validated against a dense O(N³) Gaussian process, generalised least squares, closed-form kernels and statsmodels fixtures; see doc/validation.md.