For large number of variables, it is more efficient to compute the dependence matrix separately for each state in a worker process and aggregate the results in the engine. Note that bayeslite will not be able to use this optimisation since sqlite does not support bulk pairwise, but the API should still provide such an implementation for non-bayeslite workflows.
For large number of variables, it is more efficient to compute the dependence matrix separately for each state in a worker process and aggregate the results in the engine. Note that bayeslite will not be able to use this optimisation since sqlite does not support bulk pairwise, but the API should still provide such an implementation for non-bayeslite workflows.