run_inference_algorithm can optionally take a position or an initial state. The try-except handler is a little unreliable (e.g. except not thrown in presence of other exceptions). Moreover, it seems more modular to delegate the transformation from initial_position to initial_state to the caller of run_inference_algorithm.
run_inference_algorithm produces n samples. For high dimensional problems this is memory inefficient.
transform currently only applies to state and not Info, so there isn't a way to dispense of a part of the diagnostic info.
Desired behavior
Make run_inference_algorithm only take initial_state
Allow run_inference_algorithm to have a memory-efficient mode, where it computes a running average of a desired expectation.
Change transform to take also Info as argument. This will be a breaking change (more so than (1))
Current behavior
run_inference_algorithmcan optionally take a position or an initial state. The try-except handler is a little unreliable (e.g.exceptnot thrown in presence of other exceptions). Moreover, it seems more modular to delegate the transformation from initial_position to initial_state to the caller ofrun_inference_algorithm.run_inference_algorithmproducesnsamples. For high dimensional problems this is memory inefficient.transformcurrently only applies tostateand notInfo, so there isn't a way to dispense of a part of the diagnostic info.Desired behavior
Make
run_inference_algorithmonly takeinitial_stateAllow
run_inference_algorithmto have a memory-efficient mode, where it computes a running average of a desired expectation.Change
transformto take alsoInfoas argument. This will be a breaking change (more so than (1))