Maximum size of a GM inferable by AStar

[Oliver Mader]

I tried to apply the AStar inference strategy to a problem of mine: A simple adder GM with 12 nodes and 15 states each. I can solve this very efficiently with LazyFlipper, showing only very few approximation errors.

Now I was wondering, whether AStar is also applicable to this kind of problem. The documentation states that the "graphical model must be small", but what is small? I tried the strategy with heuristic = 'fast' and a proper obectiveBound (note the spelling error), but it never terminates. I was also wondering was maxHeapSize was about, so I set it to a very small size of 100, but it also didn't help.

So basically, is my GM to large for AStar? If not, what might hinder it to infer from it properly?

Thanks,
Oliver Mader

[Oliver Mader]

Maybe a follow-up question since here. Is this algorithm in active use by anyone? If so, what type of applications, i.e., configurations in terms of search space complexity, are you working?