network too complex to be updated

[charlie77]

I have a BN with 38 nodes, all being noisy-adders. Many nodes have 4 to 5 parents and 4 states. To perform sensitivity analysis I have to convert the BN into cpt nodes, but then upon running the BN it shows "Error(42): UpdateBeliefs failed - network structure is too complex". Is such a BN really too complex? The sensitivity analysis still runs but should I trust the results if the BN cannot be updated properly?

[charlie77]

It runs when I have set the evidence for all the final parent nodes. Don't know why. The network is definitely not complex.

[shooltz[BayesFusion]]

The network can be densely connected and triangulation can fail to produce a reasonably sized jointree, even with 38 nodes.

Can you share your model, so we can investigate? You can send me a forum private message if you don't want the network to be publicly available.

[charlie77]

Hi Shooltz

Thanks for the advice. It's working as long as I set evidence to the last row of parental nodes - that's totally fine with us as we need to do that anyway.

Regards
Charles