Can SMILE be used for structure learning of a dynamic bayesian network? The tutorials only cover parameter learning.
*EDIT*
I noticed that there is an older forum post which covers this topic. Apparently SMILE does not directly support learning DBN structure. Is there any way around this limitation? (ie, force time slice 0 nodes to not have parents & not connect to one another?) Any plans to support this in the future?
Related: Would setting the "Tier" value for time slice 0 values to '1' and time slice 1 values to '2' make it work? I am not clear on what tier is.
Thank you!
Structure Learning of Dynamic Bayesian Network
Structure Learning of Dynamic Bayesian Network
Last edited by timw on Mon Jun 02, 2014 4:50 pm, edited 1 time in total.
-
- Site Admin
- Posts: 1417
- Joined: Mon Nov 26, 2007 5:51 pm
Re: Structure Learning of Dynamic Bayesian Network
The background knowledge tiers are used to block arc creation during structure learning - no arc is allowed between nodes in tiers tA and tB, where B > A. This feature won't help you in learning DBN structure, I believe.Related: Would setting the "Tier" value for time slice 0 values to 1 and time slice 1 values to 2 make it work? I am not clear on what tier is.
AFAIK, the DBN structure learning remains a research topic; we don't have any plans to incorporate such functionality into SMILearn at this point.
Re: Structure Learning of Dynamic Bayesian Network
Thanks.
I guess my best bet is to block all of the arcs I do not want to allow and then treat it like a normal BN. A little tedious, but it *might* work.
I guess my best bet is to block all of the arcs I do not want to allow and then treat it like a normal BN. A little tedious, but it *might* work.