Hi list,
I think it is due to the combined effect of:
1. the randomly restarted maximum likelihood estimation that does not start
from the same seed when the order is changed (hence the difference in the
results).
2. the flatness of the likelihood function when adding a nugget as it is
done in the
Weird, I can actually reproduce this.. Running example as is, it's fine..
but if I separately run the two parts,
I get: http://oi45.tinypic.com/1gnc5h.jpg
I'll have a look
--
Everyone hates slow websites. So do we.
Make yo
If one runs the noisy example in the basic Gaussian Process regression
example code [1] _before_ the non-noisy example then the noisy example
prediction is just a straight line. There seems to be something oddly
statefull about the GaussianProcess class.
Any thoughts on why?
Cheers,
Will
[1]
