Answered by:
Simple Hidden Markov Model
Question

Hi,
I've been trying to implement a very simple hidden markov model to understand the general concepts of how they are implemented in infer.net with c#.
I looked at the examples in Model and Model1 as well as the f# Version but I still don't get it.
Can anyone provide a very minimalistic example, for example for a hidden markov model consisting only of 3 timesteps for one site with 3 different states?
A simple Markov1 Model with only the direct predecessor influencing the next state of a site would be perfect, I added an example for the graphical model as a picture.
Just something really easy to allow me to get the general concept behind the design.
Thanks in Advance.
Monday, June 18, 2012 9:39 AM
Answers

Did you check out the WetGrassSprinklerRain example? It demonstrates how to connect discrete variables with multiple states, and how to learn the CPTs. To make a long chain, you can repeatedly invoke AddChildFromOneParent, passing in the same CPT each time.
 Marked as answer by XardasM Thursday, June 21, 2012 9:05 AM
Wednesday, June 20, 2012 4:42 PMOwner
All replies

Even a simple example that is not a HMM would be very much appreciated, maybe that would already help me to get the basics:
Lets assume I have one discrete variable with 3 states (A,B,C)
The state is influenced by the variables state in the last timestep, there is a certain probability for a statechange to either B or C, if the node is currently in state A.
I have data for a number of state changes and want to find out the probabilities for all possible state changes.
How would I do this?
Tuesday, June 19, 2012 12:42 PM 
Did you check out the WetGrassSprinklerRain example? It demonstrates how to connect discrete variables with multiple states, and how to learn the CPTs. To make a long chain, you can repeatedly invoke AddChildFromOneParent, passing in the same CPT each time.
 Marked as answer by XardasM Thursday, June 21, 2012 9:05 AM
Wednesday, June 20, 2012 4:42 PMOwner 
This is pretty much exactly what I was looking for.
I saw the example at some point, but didn't realize that the sourcecode was available as a solution.
Thanks!
Thursday, June 21, 2012 9:06 AM