How to implement a tool like MSBNx using infer.net? RRS feed

  • Question

  • hi, I want to implement a tool like msbnx (http://research.microsoft.com/en-us/um/redmond/groups/adapt/msbnx/), it allow user to build any graphical BN model(may be in a very large scale, 1000 nodes and more), and can save/load the model as a file in specific fomat, and allow user define CPTs and node probability, and then calculate the complete BN.  

    The most similiar example is the Wet grass/sprinkler/rain example(http://research.microsoft.com/en-us/um/cambridge/projects/infernet/docs/Discrete%20Bayesian%20network.aspx) , but the sample code is hardcoded

    and my question is how to start the task in a feasible way using infer.net? Is that possible? how to consider the efficiency of  calculation?


    Wednesday, January 15, 2014 3:05 AM

All replies

  • The inference algorithms used by MSBNx are not included in Infer.NET so you cannot exactly duplicate it.  In particular, MSBNx uses exact inference while Infer.NET does not.  However, you can create the same graphical models in Infer.NET and apply the Infer.NET inference algorithms, as shown in the example.
    Wednesday, January 15, 2014 2:36 PM