mogp example in Surrogates.jl

surrogates-example.ipynb

@sdwfrost
A new surrogate - AbstractGPSurrogate - has been added to surrogates.jl.

The following code where we replace the Kriging surrogate with AbstractGPSurrogate:

using AbstractGPs #necessary to define kernel - GP(SqExponentialKernel())

gp = AbstractGPSurrogate(simulation_points, simulation_output, gp=GP(SqExponentialKernel()), Σy=0.01)
@time gp_output = gp.(simulation_points);
scatter(simulation_output,gp_output,xlabel="Observed",ylabel="Predicted",legend=false,title="Training")
n_valid = 10
validation_points = sample(n_valid, lower_bound, upper_bound, LatinHypercubeSample());
validation_output = simulator.(validation_points);
@time predictions = gp.(validation_points);
scatter(validation_output,predictions,xlabel="Observed",ylabel="Predicted",legend=false,title="Testing")
n_predict = 10000
prediction_points = sample(n_predict, lower_bound, upper_bound, LatinHypercubeSample());
prediction_output = gp.(prediction_points);
prediction_se = std_error_at_point.(gp, prediction_points);
prediction_implausibility = abs.(prediction_output .- 2000)./(prediction_se .+ sqrt(400));
sum(prediction_implausibility .> 3.0)
scatter(prediction_points,markersize=5.0,group=prediction_implausibility .< 3.0)

produces a plot like this: