gaussian processes

84 views
Skip to first unread message

Matthew Pocock

unread,
Jan 15, 2015, 6:54:08 PM1/15/15
to scala-...@googlegroups.com
Hi,

I have been looking around for breeze code that implements Gaussian processes but couldn't find anything. The ConjugateGradient class looks like it will be useful for fitting the hyperparameters to training data, but I'm not quite sure how to go about it.

Thanks,
Matthew

--
Dr Matthew Pocock
Turing ate my hamster LTD

Integrative Bioinformatics Group, School of Computing Science, Newcastle University

skype: matthew.pocock
tel: (0191) 2566550

David Hall

unread,
Jan 15, 2015, 9:01:56 PM1/15/15
to scala-...@googlegroups.com
yeah there's definitely nothing for that.

--
You received this message because you are subscribed to the Google Groups "Scala Breeze" group.
To unsubscribe from this group and stop receiving emails from it, send an email to scala-breeze...@googlegroups.com.
To post to this group, send email to scala-...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/scala-breeze/CAHQS_ezfcdKBz-JH5Exf%2Bq6t-C3OVvFmUcYPyX5ShB118G-HPQ%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

Daniel Korzekwa

unread,
Jan 28, 2015, 6:14:03 PM1/28/15
to scala-...@googlegroups.com
It turns out that I actually use LBFGS from breeze for learning hyperparameters in Gaussian Processes.

Two examples:

1)Standard approach fro learing Gaussian Process Regression (marginal likelihood optimisation)

2)Learning Gaussian Processes with non gaussian likelihood (maximise variational lower bound of marginal likelihood, algortihm known as EM-EP)

David Hall

unread,
Jan 28, 2015, 6:33:03 PM1/28/15
to scala-...@googlegroups.com
awesome!

--
You received this message because you are subscribed to the Google Groups "Scala Breeze" group.
To unsubscribe from this group and stop receiving emails from it, send an email to scala-breeze...@googlegroups.com.
To post to this group, send email to scala-...@googlegroups.com.
Reply all
Reply to author
Forward
0 new messages