clvkit.CohortSurvival#

class CohortSurvival(transaction_model=None)[source]#

Bases: object

Survival curves for a non-contractual base, from a fitted P(alive).

>>> curve = CohortSurvival().fit(cb).predict()
>>> curve.to_pandas()["survival"]
>>> curve.plot()

Defaults to BGNBD(); pass any model answering fit and probability_alive to swap it, e.g. CohortSurvival(MBGNBD()).

Parameters:: transaction_model (TransactionModel | None)

fit(cb)[source]#

Fit the transaction model on cb — everything else is aggregation.

Parameters:: cb (CustomerBase)
Return type:: CohortSurvival

predict(*, period=None)[source]#

Aggregate the fitted P(alive) into a survival curve.

period is the cohort grain, any pandas offset alias (“M”, “Q”, “Y”). It defaults to the base’s own time_unit, which is the finest grain the RFM summary can resolve — coarsen it when that leaves too many cohorts to read (see opinions.md).

Parameters:: period (str | None)
Return type:: SurvivalCurve