I am using pyro’s GPRegression. I tried to examine what happens if I increase the true noise in observations and plotted an extreme example, see figure below. How come the GP reduces the assumed noise, when it is actually high? My expectation was that the standard deviation would increase.
