stopping rule for sampling methods

The front end.
Post Reply
Maral Arastoo
Posts: 2
Joined: Wed Nov 27, 2019 6:17 am

stopping rule for sampling methods

Post by Maral Arastoo »

Hello,
I would be grateful if you could help me with some questions.
I have built a massive network in GeNie. The network is too complex to be handled by exact inference methods so I am using sampling methods to perform inference. My problem is that I do not know how many samples are enough to get a good answer. Is there any stopping-rules implemented in GeNie?
Also, I have read in the manual that GeNie calculates the amount of memory needed for inference and if there is not enough memory it returns "Error(-42)". Is there any way for the user to find out how much memory is needed (even approximately)?
Best regards,
Maral Arastoo
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: stopping rule for sampling methods

Post by shooltz[BayesFusion] »

Regarding error -42, currently there's no way to obtain an estimate on memory required - it's likely to be very large anyway.

Before switching to sampling algorithm, try setting explicit target nodes in your network. This may reduce the complexity for inference algorithm.

If you're still getting error -42, even with targets, switch to EPIS inference. To set the number of samples, use Network Properties|Sampling page. The running time is roughly proportional to the number of samples.
Maral Arastoo
Posts: 2
Joined: Wed Nov 27, 2019 6:17 am

Re: stopping rule for sampling methods

Post by Maral Arastoo »

Thank you for your prompt reply.
I have set the target nodes but it still does not work.
But how can I asses the performance of the sampling methods? How many samples are needed given an accepted error? Are there any stopping rules implemented in GeNIE?
Thank you.
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: stopping rule for sampling methods

Post by shooltz[BayesFusion] »

The 'stopping rule' is just a number of samples which you can set in Network properties|Sampling.
Post Reply