search

j-faria / EPRVlnZ

MIT License
0 stars 0 forks source link

Speed up the GPs #1

Open j-faria opened 7 years ago

j-faria commented 7 years ago

Because the hyperparameters are fixed, there might be a way to speed up the matrix inversion. The problem is that the extra jitter still changes..

eggplantbren commented 7 years ago

They want a simple exponential kernel for the noise, right? We could implement that in a 'time-series' type way rather than using the GP formalism. The cost for that should be O(N).

On Wed, Jul 26, 2017 at 3:18 AM, João Faria notifications@github.com wrote:

Because the hyperparameters are fixed, there might be a way to speed up the matrix inversion. The problem is that the extra jitter still changes..

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/j-faria/EPRVlnZ/issues/1, or mute the thread https://github.com/notifications/unsubscribe-auth/ABgVOtNNNYk5eNzpWAt0SbanObgnL8W4ks5sRgXtgaJpZM4Ois09 .

-- Dr Brendon J. Brewer Department of Statistics, The University of Auckland, New Zealand Ph: +64 27 500 1336 Web: http://www.stat.auckland.ac.nz/~brewer/

j-faria commented 7 years ago

No, it's a quasi periodic kernel. model I was thinking of using a matrix identity for the inverse of the sum of matrices, and save the inverse of the non-diagonal one calculated at the start