Hi, I build a Bayesian network which has so many discretized states for each node. I can save the file. But when I run my model, I got this message"Runtime Error". Would you please help to solve it? I am just really frustrated. The attachment show the detailed message. In my opinion, it seems this is about CPU problem or the software itself. Look forward to hearing any news. Thanks a lot in advance.
BTW, the network use temporal plate. I can run for 10 time steps. But if 70 time steps, the error message is given as the picture.
Cheers
Norman
Runtime Error
-
- Site Admin
- Posts: 1417
- Joined: Mon Nov 26, 2007 5:51 pm
Re: Runtime Error
The error you're getting is caused by failed allocation of the memory required to solve the 70-step network.BTW, the network use temporal plate. I can run for 10 time steps. But if 70 time steps, the error message is given as the picture.
If you're interested only in posteriors of subset of nodes, try marking them as targets. This may (but does not have to) reduce the complexity of the unrolled network. If this doesn't help you may use one of GeNIe's sampling inference algorithms like EPIS.
-
- Posts: 4
- Joined: Sat May 08, 2010 7:53 am
Re: Runtime Error
Thank you for your reply. Is there any way to allocate enough memory so that 70 time steps can be calculated with inference algorithm "clustering" ?
Regards
Regards
-
- Site Admin
- Posts: 1417
- Joined: Mon Nov 26, 2007 5:51 pm
Re: Runtime Error
Probably not without setting explicit targets and/or reducing the state counts in nodes.normanwang wrote:Thank you for your reply. Is there any way to allocate enough memory so that 70 time steps can be calculated with inference algorithm "clustering" ?
-
- Posts: 4
- Joined: Sat May 08, 2010 7:53 am
Re: Runtime Error
Oh, that is really terrible for me !!! If I reduce the state counts, the results accuracy cannot be guaranteed. I also have tried with other algorithms, the results are not accurate enough. I think I need to compromise the both. Thanks a lot, anyway.