Sobolev Training with Approximated Derivatives for Black-Box Function Regression with Neural Networks

Repository with exemplary code of the method described in [1]. Start the example with "python3 example.py"

Description of the classes:

• The Data_Handler class provides functions for creating the data sets for training and to calculate the approximated derivatives
• The Training class provides the training functions for the neural network (which is also initialized in this function); Moreover functions for plotting the results are provided
• The function_definitions file contains the definitions of the test regression functions

Tested with Ubuntu 18.04.2 LTS; Dependencies: tensorflow, numpy, math, scikit-learn, matplotlib

[1] Kissel, Diepold: Sobolev Training with Approximated Derivatives for Black-Box Function Regression with Neural Networks