Error when running Cooper's algorithm

The front end.
Post Reply
Kazu
Posts: 10
Joined: Fri Mar 28, 2008 2:45 pm

Error when running Cooper's algorithm

Post by Kazu »

Dear all,

I have the error message shown in the subject when I tried to analyze my influence diagram.

The summary of my influemce diagram is:

Object Count States Parameters
Nodes 2119 3890 126152
Chance 1945 3890 125804
Value 174 - 348
Arcs 2646

In which situations does this error message occur?

Best regards,
Kazu
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: Error when running Cooper's algorithm

Post by shooltz[BayesFusion] »

The most likely reason is the complexity of the network. The ID influence algorithm converts ID to BN, then performs BN inference. If resulting BN network is very complex, then it may not be possible to successully perform BN inference.
Kazu
Posts: 10
Joined: Fri Mar 28, 2008 2:45 pm

Post by Kazu »

Thank you very much for the propt reply.

Actually I understand that my ID is complex...

I would like to know exactly the conditions that produce this error message. how can I see how close I am to the limitations?

With kind regards,
Kazu
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Post by shooltz[BayesFusion] »

Kazu wrote:I would like to know exactly the conditions that produce this error message. how can I see how close I am to the limitations?
This message is emitted when BN inference fails inside Cooper algorithm. It's not possible to predict "how close" you're to successful BN inference given the number of arcs/nodes/parameters.

You may consider switching BN inference algorithm to some form of sampling, then running ID inference. The ID algorithm will use the BN algorithm you selected.
Kazu
Posts: 10
Joined: Fri Mar 28, 2008 2:45 pm

Post by Kazu »

Thank you! I can imagine you cannot predict how I am close to the limitation.

But, probably you can give me the conditions implemented in the code under which the error message occur. Am I right?

Kazu
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Post by shooltz[BayesFusion] »

Kazu wrote:But, probably you can give me the conditions implemented in the code under which the error message occur. Am I right?
The condition is not expressed as explicit function of node/outcome/arc count. If you're running default BN inference (clustering) from within Cooper's algorithm, then error you're experiencing means that creation of jointree failed due to excessive size of its cliques. This condition is checked in the triangulation phase.
nzsk
Posts: 24
Joined: Thu Aug 29, 2019 8:27 am

Re: Error when running Cooper's algorithm

Post by nzsk »

Is there any particular alternative algorithm that you would recommend to switch to or try instead?
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: Error when running Cooper's algorithm

Post by shooltz[BayesFusion] »

EPIS would be the best candidate if jointree/clustering fails.
nzsk
Posts: 24
Joined: Thu Aug 29, 2019 8:27 am

Re: Error when running Cooper's algorithm

Post by nzsk »

Thank you, that solved my problem!
Post Reply