jamesrobertlloyd / gpss-research

Kernel structure discovery research code - likely to be unstable
MIT License
186 stars 73 forks source link

Fourier kernel not initisliased properly and numerically unstable #40

Closed jamesrobertlloyd closed 10 years ago

jamesrobertlloyd commented 10 years ago

Lengthscale parameter is invariant to scale - relative to period - default value of zero makes sense - also check that no out of bounds checks are made

It is also numerically unstable - past a certain lengthscale the function might as well just call CovCos

jamesrobertlloyd commented 10 years ago

https://github.com/jamesrobertlloyd/gpss-research/commit/7183e3b37480cdc9d165d25fca07075b3107c1a9

Beyond lengthscale of exp(4) it returns cosine - gradient based optimisation gets stuck but this isn't so bad - if it has got so close to being sinusoidal it might as well stay there