locked
DifficultyAbility crashing when numChoices = 2 RRS feed

  • Question

  • Thanks for such a great tool! It works great, but I've found one problem. For the ability/difficulty example, when I set numchoices = 2, it crashes since its dividing by 0. The exception is exactly: "improper distribution during inference (Gaussian(m/v=-1.285, 1/v=0)).  Cannot perform inference on this model." What could I do to resolve this? Thanks in advance!
    Thursday, April 17, 2014 1:35 AM

All replies

  • This is a known problem, and will be fixed in the next version.  Thanks!
    Thursday, April 17, 2014 9:28 AM
    Owner
  • If you want to resolve this in the current version can slightly decrease the prior variance on the discrimination variable to fix this in the current version (keep the mean the same - remember that mean is shape * scale, and variance is shape * scale^2). So for example:

    Gamma discriminationPrior = Gamma.FromShapeAndScale(6.0, 5.0 / 6.0);
    

    John

    Thursday, April 17, 2014 10:22 AM
    Owner