Parameter estimation

Parameter estimation

The sktime.param_est module contains parameter estimators, e.g., for seasonality, and utilities for plugging the estimated parameters into other estimators. For example, seasonality estimators can be combined with any seasonal forecaster to an auto-seasonality version.

All parameter estimators in sktime can be listed using the sktime.registry.all_estimators utility, using estimator_types="param_est", optionally filtered by tags. Valid tags can be listed using sktime.registry.all_tags.

A full table with tag based search is also available on the Estimator Search Page (select “parameter estimator” in the “Estimator type” dropdown).

Parameter estimators

Composition

ParamFitterPipeline(param_est, transformers)

Pipeline of transformers and a parameter estimator.

FunctionParamFitter(param, func[, kw_args, ...])

Constructs a parameter fitter from an arbitrary callable.

PluginParamsForecaster(param_est, forecaster)	Plugs parameters from a parameter estimator into a forecaster.
PluginParamsTransformer(param_est, transformer)	Plugs parameters from a parameter estimator into a transformer.

Naive parameter estimators

FixedParams(param_dict)

Dummy parameter estimator that writes fixed values to self.

Seasonality estimators

SeasonalityACF([candidate_sp, p_threshold, ...])	Find candidate seasonality parameter using autocorrelation function CI.
SeasonalityACFqstat([candidate_sp, ...])	Find candidate seasonality parameter using autocorrelation function LB q-stat.
SeasonalityPeriodogram([min_period, ...])	Score periodicities by their spectral power.

Stationarity tests

StationarityADF([p_threshold, maxlag, ...])	Test for stationarity via the Augmented Dickey-Fuller Unit Root Test (ADF).
StationarityKPSS([p_threshold, regression, ...])	Test for stationarity via the Kwiatkowski-Phillips-Schmidt-Shin Test.
StationarityADFArch([lags, trend, max_lags, ...])	Test for stationarity via the Augmented Dickey-Fuller Unit Root Test (ADF).
StationarityDFGLS([lags, trend, max_lags, ...])	Test for stationarity via the Dickey-Fuller GLS (DFGLS) Unit Root Test.
StationarityPhillipsPerron([lags, trend, ...])	Test for unit root order 1 via the Phillips-Perron Unit Root Test.
StationarityKPSSArch([lags, trend, p_threshold])	Test for stationarity via the Kwiatkowski-Phillips-Schmidt-Shin Unit Root Test.
StationarityZivotAndrews([lags, trend, ...])	Test for stationarity via the Zivot-Andrews Unit Root Test.
StationarityVarianceRatio([lags, trend, ...])	Test for stationarity via the variance ratio test for random walks.
BreakvarHeteroskedasticityTest([...])	Variance break test for heteroskedasticity.

Lag and autocorrelation estimators

ARLagOrderSelector(maxlag[, ic, glob, ...])	Estimate optimal lag order for autoregressive models using information criteria.
AcorrLjungbox([lags, boxpierce])	Ljung-Box test of autocorrelation.

Residual tests and estimators

JarqueBera([axis])

Jarque-Bera test of normality.

Cointegration

JohansenCointegration([det_order, k_ar_diff])

Test for cointegration ranks/relationships for VECM Time-Series.

Impulse and Shock Response Estimation

ImpulseResponseFunction([model, steps, ...])

Calculation of Impulse Response Parameters for various time-series forecasters.