specified with n_assimilation_batches=n on initialization of DataAssimilator
on every DataAssimilator.update() call it will create a random partition of the observed states into n evenly sized (roughly) batches, then performs EAKF updates to each batch separately.
This does not touch model parameters. Due to this, we advise only batching for the perfect data observations.
specified with
n_assimilation_batches=non initialization ofDataAssimilatoron every
DataAssimilator.update()call it will create a random partition of the observed states intonevenly sized (roughly) batches, then performs EAKF updates to each batch separately.This does not touch model parameters. Due to this, we advise only batching for the perfect data observations.