This code is part of my master thesis where I look into the use of neural networks as (hyper) potential energy surface (PES) in molecular dynamics (MD). Some simple examples, like approximating a "completely" arbitrary function of 1 or 2 variables is included and was used for training my own neural network i.e. my brain to work with TensorFlow.
Some examples of how the neural network approximates a function of two variables can be seen in the videos below: