Greykite models and components

See Choose a Model for model options and their comparison.

In Configure a Forecast, follow Choose a Model Template to set config.model_template to use the proper model.

Silverkite

Silverkite is a forecasting algorithm developed by LinkedIn.

It works by generating basis functions for growth, seasonality, holidays, etc. These features, along with any regressors you provide, are used to fit the timeseries.

The features can be combined with interaction terms in a flexible and powerful way.

This approach has the following advantages.

Flexible. Supports different kinds of growth, interactions, and fitting algorithms.

Intepretable. The default fitting algorithms are additive so you can identify the contribution of each component.

Fast. Runs much faster than Bayesian alternatives.

Prophet

Prophet is a forecasting algorithm developed by Facebook.

Details on Prophet.

ARIMA

ARIMA is a classic forecasting algorithm. We use the implementation from pmdarima.

Details on ARIMA.

model_components

config.model_components_param is an instance of ModelComponentsParam.

The attributes represent different categories of tuning parameters. See the rest of this section for how to configure each component.

from greykite.framework.templates.autogen.forecast_config import ModelComponentsParam

model_components = ModelComponentsParam(
    growth=growth,
    seasonality=seasonality,
    events=events,
    changepoints=changepoints,
    regressors=regressors,
    lagged_regressors=lagged_regressors,
    autoregression=autoregression,
    uncertainty=uncertainty,
    custom=custom,
    hyperparameter_override=hyperparameter_override
)