Derivative of Gaussian Process

[samlevy]

I am sorry to interfere, but the case you just described (1d function, with derivative observations + hierarchical model + some monotonicity and/or sign constraints) is exactly the situation I would like to learn more as well. I have a non-Gaussian likelihood with joint GP and derivative GP prior on a function. The likelihood requires the evaluation of both a function and its derivative evaluation, which motivates the use of this joint GP prior. These GP functions should be hierarchically built, as I only observe about 5-50 points per group, potentially leveraging hierarchy through the hyperparameters and/or the mean of the GP. Right now, the posterior is inferred without constraints on the GP, but I would like to impose constraints as it is consistent with theory. If the approach in the paper on rock art paintings is not recommended, I would like to learn what would be recommended. Thanks!