Implement forecast decomposition for Prophet

[brsnw250]

🚀 Feature Request

Implement forecast decomposition for ProphetModel

Proposal

In _ProphetAdapter implement method

def predict_components(df: pd.DataFrame) -> pd.DataFrame

Details:

• implement in-sample and out-of-sample decomposition in predict_components
• obtain trend component
• obtain holidays component
• seasonal components should be decomposed down to individual components (e.g. daily, weekly, yearly, custom period)
• obtain components for each exogenous variable
• raise error if model contains multiplicative components
• use target_component_ as prefix for component names

Test cases

For _ProphetAdapter

• Test components sum up to target for different parameter sets (parameters to test: holidays, growth, *_seasonality, additional_seasonality_params)
• Test with linear and logistic growth
• Test with custom seasonalities
• Test that components for exogenous variables are estimated if such variables are presented at fit stage
• Test holidays component estimated if holidays data is presented at fit stage
• Test that the component names are correct for all possible parameters combination
• Test that model decomposition contains only additive components and raises corresponding error otherwise

Additional context

No response