src.variants.metamodels.kriging module

This module is responsible for training and inference of gaussian processes (kriging).

It includes the two functions train_gp_model and infer_gp_model for training and inference. See details in their own documentations.

Contact person: Stefan Riedmaier Creation date: 04.06.2020 Python version: 3.8

src.variants.metamodels.kriging.infer_gp_model(x, model)

This function predicts outputs with a gaussian process (GP) model for the inputs x.

Parameters:

  • x (np.ndarray) – inputs of the test data set

  • model – gaussian process (GP) model

Returns:

predicted outputs of the test data set

src.variants.metamodels.kriging.train_gp_model(x, y)

This function trains a gaussian process (GP).

It is currently based on the following link using sklearn: https://towardsdatascience.com/quick-start-to-gaussian-process-regression-36d838810319

Alternatives: - GPy: often mentioned https://sheffieldml.github.io/GPy/ - GPflow: GPy fork using tensorflow https://github.com/GPflow/GPflow - PyMC3: GP with theano https://docs.pymc.io/ - excellent tutorial including numpy, sk-learn, GPflow, PyMC3 https://blog.dominodatalab.com/fitting-gaussian-process-models-python/

Parameters:

  • x (np.ndarray) – inputs of the training data set

  • y (np.ndarray) – outputs of the training data set

Returns:

trained gaussian process