locked
Modeling Additive Arrays RRS feed

  • Question

  • Question for some of you more experienced with Bayesian Nets:

    I'm trying to model an array of values that is updated constantly by adding or subtracting data.

    For example, without any machine learning, the update would look something like this with a multi-dimensional array:

    array[attribute1, attribute2, attribute3, attribute4, 0] += amount[0];

    array[attribute1, attribute2, attribute3, attribute4, 1] += amount[1];

    ect...

    Amount can be real number, positive or negative, although it is usually bound by some min/max.

    So, what I'd like to be able to do with with my Infer.NET is something like this:

    output = bayesian_model(attributes[1..4]);

    output[0] += amount[0];

    output[1] += amount[1];

    etc...

    model.train(attributes[1..4], outputs[1..n])

    Make sense?

    Sunday, May 5, 2013 5:19 AM

All replies

  • Where is the uncertainty in this problem?  So far you've only described a deterministic computation.  What are you trying to infer?

    Tuesday, May 7, 2013 12:02 PM
    Owner
  • Hey, ML_GOI here, lost my old account info.

    In response to your question: It's a deterministic computation that I'm trying to model with a Bayesian inference for the purpose of reducing memory requirements.  This is not just one array, but thousands.  As the model is trained, the new data will be an inference of the old.  It's essentially a density function.  Forgive me if my description of the problem doesn't make sense, I was hoping the pseudo code I provided would clear up what I'm trying to do.

    Anybody know how to do this?

    Friday, May 10, 2013 7:00 PM
  • It's still not clear what you are trying to do.   Try reading some of the Infer.NET tutorials and state your problem according to the patterns there.
    Friday, May 10, 2013 7:19 PM
    Owner
  • It's pretty simple, I'm trying to model an accumulation, starting at zero, then going to +/- infinity. 

    Sunday, May 12, 2013 11:33 PM
  • I also don't understand the problem. Can you express it in a factor graph? Is there a paper on this?
    Saturday, May 18, 2013 5:57 PM