locked
multivariate gaussian approximation (Migrated from community.research.microsoft.com) RRS feed

  • Question

  • nTH posted on 04-22-2010 11:46 AM

    The sample gaussian approximation that is created in the Bayes Point example...Is this the result of the inner product of w.x?  So this gaussian basically tells us which classifications are labeled True and which are labeled False?

    Friday, June 3, 2011 5:41 PM

Answers

  • John Guiver replied on 04-23-2010 12:27 PM

    I'm not sure exactly what you're asking here. The Bayes Point Machine example defines a model which has a unknown parameter vector w which has a vector Gaussian prior. We try to explain the observed data by calculating the inner product with an input vector and assigning one class ('true') if it is greater than 0, and the other class ('false') if it is less than zero. If we are in training mode, we observe the 'willBuy' value for each datum, and the inference then calculates a posterior VectorGaussian distribution for the parameter vector w. In run mode, this distribution now acts as a prior, and we infer the willBuy value.

    Also, see http://community.research.microsoft.com/forums/t/4370.aspx for a slight extension of this example which adds noise before the IsPositive factor.

    John G

    Friday, June 3, 2011 5:41 PM