DeleteNode Behavior

[yuriy]

Hello,

I'm using jSMILE so this will be in Java notation but I imagine that this is language-independent.

Suppose I have a network (of discrete CPDs) where A --> B. What happens to the CPD of B when A is deleted by calling network.deleteNode( A's handle )? It doesn't seem to be documented...

Particularly, I'm trying to get a subnetwork out of a larger network by deleting nodes and I'm wondering if the probabilities will end up consistent.

Thanks.

[shooltz[BayesFusion]]

Suppose I have a network (of discrete CPDs) where A --> B. What happens to the CPD of B when A is deleted by calling network.deleteNode( A's handle )? It doesn't seem to be documented...

The B's CPT will shrink and the remaining entries are values associated with A's 0th state. This also happens when you only delete the A-->B arc (and A remains in the network).

Particularly, I'm trying to get a subnetwork out of a larger network by deleting nodes and I'm wondering if the probabilities will end up consistent.

The probabilities managed by SMILE are always consistent.

[yuriy]

Thanks, that answers my question.

The probabilities managed by SMILE are always consistent.

By "consistent" you mean summing to 1?
I was being too vague in my last post: what I meant by "consistent" is whether P(B=1) in the resulting network will be the same as P(B=1) in the original.
Based on your response it will not.

[shooltz[BayesFusion]]

By "consistent" you mean summing to 1?
I was being too vague in my last post: what I meant by "consistent" is whether P(B=1) in the resulting network will be the same as P(B=1) in the original.
Based on your response it will not.

Yes, that's correct. When network structure changes (including arc add/delete and parent outcome add/delete), the CPTs are modified by copying existing columns or by removing the columns.