Update belief error?

[gasparz]

Hi,

I am trying to do inference in a large network (around 6100 nodes). The network contains independent trees. I am using the default algorithm for updating the beliefs (clustering I guess).

For a tree (around 1000 nodes) the algorithm does not compute the values.
(Genie shows question mark, and in smile isValueValid returns false)

Is there some built in limitation? Is there a workaround?

Thanks,
Gapsarz

[gasparz]

attached is the xsdl file for testing

[shooltz[BayesFusion]]

Is there some built in limitation? Is there a workaround?

There's no built-in limitation, at least not something we included voluntarily. The calculation of clique potentials for your network/case loses the precision; at the end of the inference algorithm the elements of these potentials are equal to zero. The potentials are used to calculate the posteriors (they're normalized to obtain probabilities). With all zeros, the information is lost.

I don' think there's any workaround except restructuring the network or limiting the node count.