Fixing a specific parameter rather than a node in EM parameter learning process

The engine.
Post Reply
Yun
Posts: 20
Joined: Mon Oct 17, 2016 8:34 pm

Fixing a specific parameter rather than a node in EM parameter learning process

Post by Yun »

Dear SMILE creators,

Could I ask whether there is a way to fix a specific parameter in a node's CPT/distribution, rather than fixing the entire node during the EM parameter learning process? Currently, I am using "em.learn(ds, net, matching, fixedNodes)", and I found that "fixedNodes" will fix all the parameters related to a node. However, I just want to fix one of them (based on the expert knowledge). For example, if a binary node has one binary parent, then this node will have two parameters, and I only want to fix the value of one of the parameters while allowing changing the other parameter. Is there a way to do so? If not, is it possible to add this option?

Thank you so much for your help! I appreciate it!

Thanks,
Yun

Intelligent Systems Program
University of Pittsburgh
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: Fixing a specific parameter rather than a node in EM parameter learning process

Post by shooltz[BayesFusion] »

Is there a way to do so? If not, is it possible to add this option?
Unfortunately, this is not possible. We don't have plans to change this behavior, as it would require substantial amount of work and quite complex API changes.
Yun
Posts: 20
Joined: Mon Oct 17, 2016 8:34 pm

Re: Fixing a specific parameter rather than a node in EM parameter learning process

Post by Yun »

Thank you very much for letting me know! It's really hard to find toolkits that have this option, so I understand this need might be too specific. Thank you all the same!
Post Reply