pypomp.PanelPomp.mix_and_match

PanelPomp.mix_and_match() → None

Sort parameters independently and cross-combine them.

Ranks the shared parameters of all replicates based on their overall panel log-likelihoods. Independently, for each unit, ranks the unit-specific parameters of all replicates based on that unit’s individual log-likelihoods.

Finally, reconstructs the parameter set by pairing them by rank: the n-th new replicate is formed by combining the n-th best shared parameter set with the n-th best unit-specific parameter set for each unit. This cross-combines the best-performing components from different replicates to construct potentially superior new parameter sets.

In particular, if all parameter sets share the same values for the shared parameters, then the overall panel likelihood factors completely across units. Consequently, the new best replicate (rank 0) is guaranteed to have a panel log-likelihood equal to the sum of the maximum log-likelihoods obtained for each unit individually across all original replicates.