Looking to calculate evidence for a BPM Model RRS feed

  • Question

  • I'm using the provided BPM example to perform classification tests on noisy versions of standard datasets from the UCI repository for my honours project.

    However I've been trying for a while and can't seem to work out how to get the log evidence for the models being generated. I wonder if anyone had any idea of how to go about doing this? I've had a look at the cyclist examples for how their evidence values are fetched but can't seem to find how to apply this to the BPM.

    Any help/pointers would be much appreciated. 

    Many thanks,

    Guy Templeton

    Sunday, April 12, 2015 6:51 PM

All replies

  • Hi Guy,

    Computing model evidence is explained in the user guide.

    You can also look at our BPM implementation in the Learners folder of Infer.NET 2.6, which computes evidence. The code is under $\Infer.NET 2.6.zip\Learners\ClassifierModels\DenseBinaryModel.cs in the constructor of the class.


    Sunday, April 12, 2015 7:38 PM