Implement method for Gaussian process regression with binary response

[jamesotto852]

Chapter 3 of Gaussian Processes for Machine Learning outlines non-parametric methods for estimating the success probability of a spatially correlated binary response. The main idea is assuming the probability of success is a Gaussian proccess that has gone through an inverse link function. The mathematics involved are more complicated than kriging, and the simplest approaches involve Laplace approximations in Bayesian inference. It would be good to implement this in ggspatreg.

[jamesotto852]

I have done this manually before:

Here is a plot showing the estimated probability surface for some spatially correlated coin flips:

Here is another example, this time using the Ames data set. Here, we have both the GP approach and a non-spatial logistic regression.