none
Online learning where posterior factors over two indices

    Question

  • Would it be possible to do online parameter estimation with Infer.net in a situation like the StudentSkills tutorial where instead of obtaining a students response to all questions, the data comes in 1 question at a time (e.g. student 5 answers question 3, next data point is student 7 answers question 2, ...etc). 
    Wednesday, September 13, 2017 3:46 PM

Answers

  • That shouldn't be a problem.  The number of students and their priors can be observed variables.  If a new student arrives, you bump up the number by one and extend the prior array.
    • Marked as answer by cyentist Friday, September 15, 2017 3:19 PM
    Wednesday, September 13, 2017 7:07 PM

All replies

  • I don't see why not.
    Wednesday, September 13, 2017 5:03 PM
  • In the user guide it explains online learning in the context of learning a gaussian. In that case the data that streams in always relates to 1 gaussian. In my case a new data point I receive might be for a student I have  not seen before and for whom I do not have a prior setup. 
    Wednesday, September 13, 2017 6:31 PM
  • That shouldn't be a problem.  The number of students and their priors can be observed variables.  If a new student arrives, you bump up the number by one and extend the prior array.
    • Marked as answer by cyentist Friday, September 15, 2017 3:19 PM
    Wednesday, September 13, 2017 7:07 PM