Hi, Can I ask how the probability of evidence is calculated in SMILE?
As far as I know, the pe is queried by summing over all possible states which mean the space is exponential. So How Smile does that?
thanks
Peng
how probability of evidence is calculated
-
- Site Admin
- Posts: 1417
- Joined: Mon Nov 26, 2007 5:51 pm
Re: how probability of evidence is calculated
By default SMILE calculates P(e) using jointree-based algorithm; if the jointree resulting from triangulation does not fit into memory, the fallback algorithm based on chain rule is used.
You can force the DSL_network::CalcProbEvidence to use chain rule by passing true as its 2nd argument.
You can force the DSL_network::CalcProbEvidence to use chain rule by passing true as its 2nd argument.
Re: how probability of evidence is calculated
thanks for the answers.
Can I ask one more question? How to access each probability table associated with each node in the Network? I want to try to calculate the probability of evidence by chain rule but using a subset of all the probability tables.
many thanks
snowave
Can I ask one more question? How to access each probability table associated with each node in the Network? I want to try to calculate the probability of evidence by chain rule but using a subset of all the probability tables.
many thanks
snowave
-
- Site Admin
- Posts: 1417
- Joined: Mon Nov 26, 2007 5:51 pm
Re: how probability of evidence is calculated
Do you mean the conditional probabilities (node definition, stored in xdsl when you save network to disk) or calculated posterior probabilities (node value)?How to access each probability table associated with each node in the Network?
Re: how probability of evidence is calculated
yes, I mean the node definition, conditional probability. I think this can be used to calculate the probability of evidence by chain rule.
-
- Site Admin
- Posts: 1417
- Joined: Mon Nov 26, 2007 5:51 pm
Re: how probability of evidence is calculated
See the "Anatomy of the node/Node Definition" section in GeNIe manual. Tutorial 3 (Exploring the contents of a model) may also be helpful.