weighted training data for gaussian process regression

Hello everybody,

I've got a question regarding the training of a gaussian process. My training data table consists of two type of data, labelled with 'A' and 'B'. The number of rows for 'A' is quite large and makes ~90% of the complete data set, whereas the number of rows for 'B' is quite small (~10%). Now I want to increase the weight of 'B' in the training data set, so that both labels are equally represented in the model. What I don't want, is to remove data with the label 'A'.

Does anyone have an idea how to do this?

Hints, remarks, links are highly appreciated!

N.

[Jan Pecka]

Hi Nicolai, I have exactly the same problem. Have you come across an usable solution?