Chance vs Deterministic Nodes

The front end.
Post Reply
jdtoellner
Posts: 71
Joined: Mon Aug 01, 2016 9:45 pm

Chance vs Deterministic Nodes

Post by jdtoellner »

I have created a network with Chance nodes whos CPT's are all 1's and 0's.

Is there any computational advantage to using Deterministic nodes in this instance?
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: Chance vs Deterministic Nodes

Post by shooltz[BayesFusion] »

Is there any computational advantage to using Deterministic nodes in this instance?
No, there's no explicit check for node type at this point. The logic flow in the inference algorithm is based on the actual values in the CPTs, not on the node type.
jdtoellner
Posts: 71
Joined: Mon Aug 01, 2016 9:45 pm

Re: Chance vs Deterministic Nodes

Post by jdtoellner »

The same answer would then hold for Noisy-Max nodes and CPT.

If I'm creating Noisy-Max nodes they end up being represented as CPTs. There's no advantage to creating them as CPTs in the first place.

(Which means I'd go with Noisy-Max since they're so much easier to set up.)
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: Chance vs Deterministic Nodes

Post by shooltz[BayesFusion] »

If I'm creating Noisy-Max nodes they end up being represented as CPTs
They have their noisy weights expanded into full CPTs during inference.
There's no advantage to creating them as CPTs in the first place.
If noisyMax nodes are expressive enough for your purpose, go for it. Additionally, there are parts of SMILE's relevance layer which look for noisyMax nodes and take advantage of their presence to simplify the intermediate data structures created when inference runs.
Post Reply