Solution postprocessing

[riyadm]

Hi!

I was thinking it'd be useful to have a set of simple methods for solution postprocessing, e.g.

mean(solution)
var(solution)
quantile(solution, p)

Some of the other possibilities: mean above/below a threshold, probability (frequency) above/below a threshold.

These are simple, yet provide a nice and quick way of looking at a summary statistic of a solution. Some of these methods may be extended from Statistics module.

Would also be great to hear feedback from users on which methods would be useful. Thanks!

[juliohm]

That is a great idea. We can start broadening our view of a SimulationSolution as defined in GeoStatsBase.jl to be an empirical distribution in which geostatistical realizations are the samples. In this general sense, we could start computing summary statistics like the mean, var, and quantile as suggested.

I think the code should live in GeoStatsDevTools.jl because it can be useful to other packages and solvers in the future. I agree we should extend the methods in Statistics 👍

What is still unclear to me is the output format of these functions. What would be useful from a user's perspective? Let's say we take the mean of a solution, what is the format of the output? Should we return a dictionary mapping properties to mean maps? Something like the output of the digest function we already have. We could start by returning another SimulationSolution object with a single realization containing the mean, and then calling the already implemented digest on it as the return value.

If you can sketch an initial implementation of the mean function for example in a file called GeoStatsDevTools.jl/src/solutions/postprocessing.jl that would be very helpful. Feel free to add your name in the copyright header like in other files ;)

[juliohm]

Actually, you could start adding the functions to the existing file GeoStatsDevTools.jl/src/solutions/simulation_solution.jl. This is a generic place in the source where we can add functionality for SimulationSolution objects.

[juliohm]

This is being tracked in juliohm/GeoStatsDevTools.jl#2

[juliohm]

Fixed on master, thanks @riyadm.