locked
laplace distribution RRS feed

  • Question

  • I want to implement Laplace distribution and also define a factor multiply(Laplace,Gaussian). Is it possible? Can anyone let me know from where can I start? Thank you!
    Monday, October 20, 2014 3:40 PM

All replies

  • Would this help?
    Monday, October 20, 2014 10:18 PM
  • I already saw this. But this does not work for me since it is an approximation of Laplace. I need a l1-norm regularization (lasso method) which needs a Laplace prior distribution on parameter vector (to prevent overfitting). If I use an approximation of Laplace, I just get all parameters near zero (I mean it does not give a sparse model as I expect). So I think I need to implement the Laplce distribution so I can have a really Laplace prior (and not just an approximated Laplace).
    I hope my explanation is correct. I'm new in this and maybe I've got something wrong?
    • Edited by Capli19 Wednesday, October 22, 2014 4:07 PM
    Tuesday, October 21, 2014 9:56 AM