How to account for the measured uncertainties in gp.Marginal?

There’s Marginal.marginal_likelihood method that has noise argument. It is WhiteNoise covariance initialized with a scalar (variable). But what if i have measured uncertainties for each data point?

P.S. With Latent implementation it is clear how to use the measured uncertainties.

Not sure I understand, the noise argument in gp.marginal_likelihood represents the uncertainties of the measurement. If you meant each data point has a different noise (so noise is a vector same size as observed) I think it should work by passing a vector to noise.

1 Like

It works with just setting noise to an array of the measured uncertainties. Don’t know why this didn’t work earlier!

Thank you!