jamesrobertlloyd
jamesrobertlloyd
Have experimented with latest version of GPML optimiser - same local extremum problems around - asking Carl. Might be instructive to try the failure example in GPy to see if...
Gershenfeld violin + time series prediction book
Birth data from latest edition of Bayesian Data Analysis (copy has been pre-ordered)
http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4487100 Can get data via Oliver Stegle from Soren Broge
Hannes has supplied some code - saved under GPML in my gmail - to be introduced
As a temporary quick fix, the translation can just note when a transition is soft, without giving exact details
Noticed it again - need to be wary of this.
Probably not - currently turned it off for k = 1 so new experiments will generate unadulterated data
Just spoke with David - it does exist...!
Any ideas why - or is it worth others testing?