Hello,
I do not know if this group is still maintained these days, but I'll just try :).
I'm trying to use OpenGM instead of libDAI in my code. For my problem libDAI works well, but I would not mind to move to openGM as it is header only and has a few extras.
Unfortunately I do not comprehend how to get my problem working with openGM.
I have a set of nodes which are part of a factor graph. For the factor's states until now I assigned beliefs, something like e^(-distance). Distance in that case would be the energy of the system if I'm correct?
I do not understand, is it possible to assign probabilities to the DiscreteSpace's factor tables? Is it just a matter how to configure the BP later on? Or am I supposed to assign energies to each factor state?
In libDAI you can enable a property called "logdomain", and you can specify the mode as "MAXPROD". And in that case it works out of the box.
Do you have an idea of how to configure openGM's BP to work in a similar way? I tried to convert my beliefs to energies and then run BP with Minimizer, but somehow I'm not getting the correct results on my artificial example.
Thank you and kind regards,
Phil