Modification of the Wet grass/rain/sprinkler example RRS feed

  • Question

  • Hello,

    If we were to modify this example (http://research.microsoft.com/en-us/um/cambridge/projects/infernet/docs/Discrete%20Bayesian%20network.aspx) to have a latent variable, e.g. wet_grass, and if we wished to infer the parameters of the model from observations of the remaining primary variables (cloudy, rain, sprinkler), what modifications in the code would be needed?

    Setting WetGrass.ClearObservedValue() in LearnParameters(..) is not sufficient as I understand that I need to break symmetry somehow?

    Thank you!

    Monday, February 18, 2013 8:55 AM


  • I don't think any symmetry breaking is needed. If you never observe wet grass, the wet grass CPT will stay at its prior uniform value - there is no constraint to drive it from its prior. You will then also get a uniform distribution over wet grass itself.
    • Marked as answer by alex.nova Monday, February 25, 2013 9:56 PM
    Thursday, February 21, 2013 11:26 AM