Sign in
Microsoft.com
 
Microsoft
 
 
 

locked
VB lower bound in mixture of Gaussians model? RRS feed

 > 

  • Question

  • Question
    Sign in to vote
    0
    Sign in to vote

    I am trying to implement a mixture of Gaussians model. I used the provided example as a basis but I'm trying to add monitoring of the VB loglikelihood lower bound. The code is available here:

    http://pastebin.com/2sULq1hz

    I get the following error:

    Running mixture of Gaussians
Compiling model...compilation failed.

Unhandled Exception:
MicrosoftResearch.Infer.CompilationFailedException: MessageTransform failed with 1 error(s) and 0 warning(s):
Error 0: This model is not supported with VariationalMessagePassing due to Wishart.SampleFromShapeAndScale(PositiveDefiniteMatrix sample, double shape, PositiveDefiniteMatrix scale). Try using a different algorithm or expressing the model differently in
Wishart.SampleFromShapeAndScale(100.0, vPositiveDefiniteMatrix0)
    Full error message:

    http://pastebin.com/raw.php?i=esKckVQw

    If I comment out the IfBlocks (lines 34 and 69), everything works (except the lower bound computation). Any ideas how to fix this?

    Thursday, November 13, 2014 3:36 PM

All replies

  • Question
    Sign in to vote
    0
    Sign in to vote
    Use Variable.WishartFromShapeAndRate instead.
    Friday, November 14, 2014 3:11 PM
    Owner
  • Question
    Sign in to vote
    0
    Sign in to vote
    The version using WishartFromShapeAndScale should now work properly in Infer.NET 2.6.
    Saturday, November 29, 2014 6:09 PM
    Owner