Aggregated Transitions

[yannikschaelte]

When different parameter dimensions need to be treated differently, e.g. with a different bandwidth scale, or only partial multimodality, or different discrete setups, it would be good to allow the transition needed in SMC to account for this. Don't know if straightforward to do, but one could think about an AggregatedTransition, which holds multiple sub-transitions, and passes reduced input on to them, and gathers/aggregates the results, assuming independence usually.

[chrhck]

Hi, is there any progress on this issue? I have an application with both discrete and continuous priors where this would come in handy.

[yannikschaelte]

Hi @chrhck , unfortunately there has been no progress on this part. For the moment, the best way may be to use a continuous transition for all parameters, and round the obtained parameters in your model. This is theoretically not completely right, but has worked in the past.

[chrhck]

Thanks, I'll try that for now.

[yannikschaelte]

Hi @chrhck , it has been a long time unfortunately, but if this is still a problem (probably not :/) and if you have a finite discrete domain, then recent work (in development) might help with that, see the above merge.