Bayesian ets

`polars_ts.models.bayesian_ets`

Bayesian Exponential Smoothing (Bayesian ETS).

`BayesianETS`

Bayesian Exponential Smoothing forecaster.

`fit(df)`

Fit the Bayesian ETS model.

`predict(df, h)`

Generate h-step-ahead forecasts with credible intervals.

`BayesianETSResult` `dataclass`

Fitted result from BayesianETS.

`ETSPriors` `dataclass`

Prior distributions for ETS smoothing parameters and initial states.

`_forecast_from_params(values, params, model, m, additive, h, sigma_noise=False, rng=None)`

Run ETS forward to get h-step forecasts from fitted parameters.

`_holt_loglik(values, alpha, beta, level0, trend0, sigma)`

Gaussian log-likelihood for Holt's linear trend model.

`_hw_loglik(values, alpha, beta, gamma, level0, trend0, seasons0, m, additive, sigma)`

Gaussian log-likelihood for Holt-Winters model.

`_log_posterior(theta, values, model, m, additive, priors)`

Compute unnormalized log-posterior.

`_map_estimate(values, model, m, additive, priors)`

Find MAP estimate via L-BFGS-B.

`_mcmc_sample(values, model, m, additive, priors, n_samples=1000, burn_in=500, seed=42)`

Draw posterior samples via Metropolis-Hastings.

`_pack_params(model, alpha, beta, gamma, level0, trend0, seasons0, sigma)`

Pack parameters into a flat array for optimization.

`_ses_loglik(values, alpha, level0, sigma)`

Gaussian log-likelihood for SES state-space model.

`_unpack_params(theta, model, m)`

Unpack flat parameter array into named parameters.

`bayesian_ets(df, h, model='ses', inference='map', season_length=1, seasonal='additive', priors=None, coverage=0.9, n_samples=1000, burn_in=500, seed=42, target_col='y', id_col='unique_id', time_col='ds')`

Bayesian ETS convenience function.

Bayesian ets

polars_ts.models.bayesian_ets

BayesianETS

fit(df)

predict(df, h)

BayesianETSResult dataclass

ETSPriors dataclass

_forecast_from_params(values, params, model, m, additive, h, sigma_noise=False, rng=None)

_holt_loglik(values, alpha, beta, level0, trend0, sigma)

_hw_loglik(values, alpha, beta, gamma, level0, trend0, seasons0, m, additive, sigma)

_log_posterior(theta, values, model, m, additive, priors)

_map_estimate(values, model, m, additive, priors)

_mcmc_sample(values, model, m, additive, priors, n_samples=1000, burn_in=500, seed=42)

_pack_params(model, alpha, beta, gamma, level0, trend0, seasons0, sigma)

_ses_loglik(values, alpha, level0, sigma)

_unpack_params(theta, model, m)

bayesian_ets(df, h, model='ses', inference='map', season_length=1, seasonal='additive', priors=None, coverage=0.9, n_samples=1000, burn_in=500, seed=42, target_col='y', id_col='unique_id', time_col='ds')