4.2.2. Model Instantiation

The class Gaussian Process Regressor (GPR) of the scikit-learn package class allows the implementation of Gaussian process models. For the instantiation in particular, a definition of a kernel was needed. This kernel is also called covariance function in connection with Gaussian processes and influences the probability distributions of the Gaussian processes decisively. The main task of the kernel is to calculate the covariance of the Gaussian process between the individual data points. Two GPR objects were instantiated with two different kernels. The first one was created with a standard kernel and the second one was additionally linked with a white noise kernel. During the later model training, the hyperparameters of the kernel were optimized. Due to possibly occurring local maxima, the passing parameter n\_restarts\_optimizer can be used to determine how often this optimization process should be run. In the case of GPR, a standardization of the data was carried out. This standardization was achieved by scaling the data mean to 0 and the standard deviation to 1.
