Currently the state/parameter fitting routines only support one scalar-valued observation per node. However, multi-strata models would like to use vector-valued observational data. For example, an age-stratified model using weekly hospitalization data would want to incorporate data for each age strata.
This change would touch a few components:
ParticleFilterSimulator and EnsembleKalmanFilterSimulator would need to support multiple observations per node.- The
ModelLink collapses the quantity dimension. Instead, this dimension would need to be preserved.
- The likelihood functions would need to support multiple observations (this is straightforward, assuming the observations are independent per strata.) We may also need to add tools for "gluing" several likelihoods together.
- The ADRIO API already supports various data shapes. For convenience we may also want the ability to stack the output of multiple ADRIOs together.
Currently the state/parameter fitting routines only support one scalar-valued observation per node. However, multi-strata models would like to use vector-valued observational data. For example, an age-stratified model using weekly hospitalization data would want to incorporate data for each age strata.
This change would touch a few components:
ParticleFilterSimulatorandEnsembleKalmanFilterSimulatorwould need to support multiple observations per node.ModelLinkcollapses the quantity dimension. Instead, this dimension would need to be preserved.