locked
Prior on precision RRS feed

  • Question

  • Hello everyone,

    In my regression problem, I need to specify a prior on precision. 

    My data is in range [0,1], so my prior on precision won't be so small. (I know data noise is less than 10 percent).

    I was looking for the right prior for my noise parameter (I mean noise or precision of the output variable, which is itself a Gaussian from unknown mean and precision). I see GammaFromShapeAndScale(1,99999) can be an uninformative prior on precision. But how about this case that I know it is less than 10 percent, how should I specify my Gamma distribution as a prior on this precision?

    Can you please help me in this case?

    Sunday, February 7, 2016 1:47 PM

Answers

  • The mean of a Gamma distribution is shape*scale, and its variance is shape*scale*scale.  You can use this to work out the desired parameters.  Alternatively, you can use the function Gamma.FromMeanAndVariance.
    • Marked as answer by RazinR Sunday, February 7, 2016 8:40 PM
    Sunday, February 7, 2016 4:49 PM
    Owner