1) What is the origin of the Kernel (RBF + Bias + Fixed) being used in the model? The Fixed part seems to rely on the covariance (I think?) matrix of the data.
2) The acquisition function has obviously been written in a very parsimonious way. Is this the Expected Improvement function? It actually looks like the GP Lower Confidence Bound. Also, why is the kernel now being used just RBF + Bias.
Hi,
A few questions regarding the GP strategy:
1) What is the origin of the Kernel (
RBF+Bias+Fixed) being used in the model? TheFixedpart seems to rely on the covariance (I think?) matrix of the data.2) The acquisition function has obviously been written in a very parsimonious way. Is this the Expected Improvement function? It actually looks like the GP Lower Confidence Bound. Also, why is the kernel now being used just
RBF+Bias.Any references (beside Snoek) would be great.
Thanks,
Rob