Skip to content

Instantly share code, notes, and snippets.

@fehiepsi

fehiepsi/gaussian.ipynb

Last active March 24, 2019 01:04
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save fehiepsi/5ef8e09e61604f10607380467eb82006 to your computer and use it in GitHub Desktop.
Save fehiepsi/5ef8e09e61604f10607380467eb82006 to your computer and use it in GitHub Desktop.
Download ZIP
Display the source blob
Display the rendered blob
Raw
gaussian.ipynb
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
@fritzo
Copy link

fritzo commented Mar 1, 2019

This is great! We should add it to funsor/derivations/gaussian.ipynb 😄

@fritzo
Copy link

fritzo commented Mar 1, 2019 •
edited

Re: normalization: In the current design, Gaussians are essentially numerically stable representations of strictly concave quadratic polynomials, i.e. they are intentionally non-normalized. Thus the log(2 pi) and log det terms appear only when reducing via .logaddexp(). This is a bit tricky since we often reduce only a subset of dims, and only that subset get the log(2 pi) terms. I don't yet know how to compute the log det terms of the partial reduction.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment