Learning on fixed datasets

[ArzhelaR]

Learning is actually performed either on a fixed mesh or on randomly generated meshes.
The goal is to enable training on a fixed dataset provided by the Tune Dataset project.