Setting observed probabilities for a node

[ausi]

Hi!
I'm using SMILE with Java and have created a Bayesian Network with some Nodes where I can set my observations as evidence and one node that gives me the result.
My Problem is, that my observations aren't exact. Evidence means that a state of a node is observed with probability 100%. In my case I'd like to set the probability of each state.

E.g. a Node with four states: state0 0%, state1 30%, state2 60%, state3 10%

Is this possible? I only found setEvidence() to put in observations, which doesn't solve my problem.

Thanks for your advice!

Ausi

[shooltz[BayesFusion]]

Is this possible?

It is not possible with current SMILE implementation.

[mark]

This thread may be of interest to you: http://genie.sis.pitt.edu/forum/viewtopic.php?t=5

A post from this thread gives an alternative solution (I don't know how to link the post directly):

By the way, to simulate soft evidence I have found a good way:

For example if I want to set 30% state1 and 70% state2. I just set 100% state1. Then I multiply the results by 0.3. After that I set 100% state2. And multiply these results by 0.7.

The adding both values and I have soft evidence. Sounds very simple Smile

Maybe I am making a big mistake in my thoughts, but if not you could very easy adapt that in smile.

[ausi]

Hi Mark!
I don't know what the guy in your quotation is talking about , but the other thread was very helpful. Thanks!

Soft Evidence is what I need and under this term I already found some good starting points on the net. Maybe I try to "simulate" soft evidence with an extra child node.
Do you know a tutorial for this by any chance?

I'm just wondering a bit, why this isn't provided by SMILE. Isn't that something needed by quite some users? Is there a chance that it will be intergrated into SMILE in the near future?

Regards, Ausi

[mark]

I don't know about a tutorial on soft evidence. I'm also not quite sure why we don't have soft evidence. I'll ask the person responsible for that decision to write down a few words in this topic.