Sign in
Microsoft.com
 
Microsoft
 
 
 

locked
Models for Cognitive diagnostic models RRS feed

 > 

  • Question

  • Question
    Sign in to vote
    0
    Sign in to vote
    I saw the presentation slides for "Automating variational inference for statistics and data mining" (IMPS 2009) which used how infer.net can be used for the DINA and NIDA models. I haven't seen any papers come out using infer.net for cognitive diagnostic models however. Is there a reason? Did EP not work particularly well in these cases? Were there other issues? Or just no interest? 
    Wednesday, September 6, 2017 2:49 PM

Answers

  • Question
    Sign in to vote
    1
    Sign in to vote
    The models used in this talk were turned into the Student skills example that comes with Infer.NET.  If you are asking why Infer.NET is not widely used in that community, I can't give a definitive answer.  I do know that there are many existing libraries for inference in cognitive diagnostic models, so that may be a factor.
    • Marked as answer by cyentist Wednesday, September 6, 2017 9:24 PM
    Wednesday, September 6, 2017 5:24 PM
    Owner

All replies

  • Question
    Sign in to vote
    1
    Sign in to vote
    The models used in this talk were turned into the Student skills example that comes with Infer.NET.  If you are asking why Infer.NET is not widely used in that community, I can't give a definitive answer.  I do know that there are many existing libraries for inference in cognitive diagnostic models, so that may be a factor.
    • Marked as answer by cyentist Wednesday, September 6, 2017 9:24 PM
    Wednesday, September 6, 2017 5:24 PM
    Owner
  • Question
    Sign in to vote
    0
    Sign in to vote

    Thanks Tom. One of the limitations of these models is that they are mixture models which marginalize out the latent class (which, if there are K skills, there are 2^K latent classes) which becomes computationally difficult (for large K) for the existing libraries to handle as they typically use EM algorithm. 

    I'm only just becoming familiar with EP. Do you know if this limitation would be a problem when fitting the model with EP as well? 

    Wednesday, September 6, 2017 7:17 PM
  • Question
    Sign in to vote
    1
    Sign in to vote
    EP does not enumerate all 2^K latent classes so it would not have that problem.
    Wednesday, September 6, 2017 7:40 PM
    Owner