Fitting a basic neural network to samples of a simple function using the tch crate. The resulting predictions for the training samples are shown in a created plot looking like the one below.
The underlying function the neural network is trained to approximate is given by 0.09 * x^3 + 2 * cos(x) - 1. During the training data creation an artificial sampling error is applied given by normal distributed error fractions.
After checking out the repository and building and running by invoking cargo run a comparable image output should be created. Be aware that building can take a while as the tch crate will depending on the exact setup download the desired version of C++ PyTorch library (libtorch).