Easy Regression Support

The concept of regression can be used to learn the relationship between input features and extract deeper features.

In the framework of the ttZ analysis these deeper features are the kinematics of the top-Z system. Instead of trying to reconstruct the particles themselves one could leave this task to a neural network. Such a regression NN would then predict the kinematics directly.

For such a feature it needs to be possible to pass float values as labels to enable regression. In addition some fail safes need to be added to ensure the conversion of the regression algorithms -> We probably need dedicated and custom-made loss functions too.