jump-start on Infer.NET bit string model (Migrated from community.research.microsoft.com) RRS feed

  • Question

  • Jack in Danville posted on 04-15-2009 11:30 AM

    I'm somewhat familiar with Bayesian Inference from the literature, but have no experience. I'm requesting someone help jump-start me on Infer.NET. Here is my problem:

    Input is long bit strings, currently in the range 32-560 bits, but can be longer.
    Input bit string consists of pairs of bits, each pair representing one of 4 possible states.
    There may or may not be a "hard for humans to discover" relationship between any of the pairs.
    The input set can be all fixed length, but I see Infer.NET does support jagged training sets, which is convenient and I may exploit once I gain experience.

    I want to map these long bit strings to a shorter bit string, typically 3-9 bits, in which one and only one bit gets set.

    The problem is discovering the mapping of new long bit strings to short strings. The training set consists of all verified mappings, and is small to begin with, but grows as new mappings are approved.

    All tips appreciated! (Especially code snippets; and it's OK to assume I'm a complete idiot and need to be walked through every step.)

    Friday, June 3, 2011 4:51 PM


All replies