In EM learning, how to keep the right relationships

The engine.
Post Reply
musicpxg
Posts: 19
Joined: Sat Feb 16, 2013 6:24 am

In EM learning, how to keep the right relationships

Post by musicpxg »

Hi,

I will show my question by an example:
For instance, I have three nodes, A, B, C
C is child of B, B is child of A.
A has two states, a0,a1
B has two states, b0,b1
C has two states, c0,c1

B is an invisible node in the training data set. Only data for A and C are available.
a0 has strong relationship with b0.
b0 has strong relationship with c0;

BUT, in the training results,
a0 has strong relationship with b1.
b1 has strong relationship with c0;

I know the result makes sense. My question is how to make it to be b0 instead of b1 in the results?

Thank you!
Best wishes,
Yang
mark
Posts: 179
Joined: Tue Nov 27, 2007 4:02 pm

Re: In EM learning, how to keep the right relationships

Post by mark »

After running EM you can swap the b0 and b1 state names.
musicpxg
Posts: 19
Joined: Sat Feb 16, 2013 6:24 am

Re: In EM learning, how to keep the right relationships

Post by musicpxg »

mark wrote:After running EM you can swap the b0 and b1 state names.
It's a good idea.

I found there is a function network.setOutcomeId(), I think it can be a solution.

Thank you!
Post Reply