I am trying to implement Bayesian Networks learning using GraphLab. In order to do that I am using the factor graphs implementation that Scott wrote, with some modifications (I turned some private methods into public). Currently I can perform evidence and inference, but I want to implement both parameter learning (for the complete data and incomplete data scenarios) and structure learning (also for complete and incomplete data).
In order to implement those learning tasks I need to be able to read data from a table. For the parameter learning I assume I already have a structure for the network and that each line of the table represents a particular assignment for the variables in the network.
I don't know if there is an easier way than the one I explained, but I believe I need, some how, to read this data from a table so I can propagate the information through the graph and do the learning.
Finally, once I finish this implementations and test it in some common test cases I am going to send it to you guys in case you want to incorporate those algorithms to the GraphLab source code.
Thank you for your time helping me :smiley: