Implementation for the first Figure of the article "Gaussian process interpolation with conformal prediction: methods and comparative analysis"
First, create a virtual environment then install the requirements. Finally, install GPMP package to build Gaussian process models.
The file notebook goldstein_price_cloud
shows how to create the points
cloud in the article for the
Godlstein-Price function.
You can create other cloud for functions using the test functions implemented in
the module gpmp.misc.testfunctions
or using the functions in
src.functions.py
.
The cloud is built using the class GPExperiment
implemented in the file
src.gpmodelmetrics.py
. When you instantiate the class, a design of
experiment is automatically created. Then you can use the two above methods:
- The
evaluate_model_variation
method is used to generate the cloud around the parameter selected by restricted maximum likelihood. - The
j_plus_gp_point
method is used to compute the IAE when the prediction interval are computed using J+GP method.