Create BaseReconciler

[alex-hse-repository]

🚀 Feature Request

Create class to perform forecast reconcilation

Proposal

Create class:

class BaseReconciler(ABC, BaseMixin):

    def __init__(self, target_level: str, source_level: str):
        """
        target_level: level to be reconciled from the forecasts
        source_level: level to be forecasted
        """
        self.mapping_matrix: Optional[scipy.sparse.base.spmatrix]

1. Add abstract method fit(self, ts: TSDataset) -> "BaseReconciler"

• Receive dataset on the the level which is lower or equal than target_level, source_level
• Aggregate the dataset to the source_level
• Fit the parameters of the Reconciler(mapping_matrix) -- will differs in children

2. Add method aggregate(self, ts: TSDataset) -> "TSDataset"

• Receive dataset on the the level which is lower or equal than source_level
• Aggregate the dataset to the source_level

3. Add method reconcile(self, ts: TSDataset) -> TSDataset:

• Receive dataset on source_level
• Reconcile the forecasts form ts using mapping_matrix to the target_level
  • Y_target = G * Y_source.T
• Create dataset with new df:

TSDataset(df=Y_target, freq=ts.freq, ...)

Test cases

1. Method aggregate work correctly for different source_level(lower/equal/greater than level of ts)
2. Method reconcile works correctly for top_down/bottom_up mapping matrixes and different level of input ts(lower/equal/greater than source_level)

Additional context

No response