WIP: Active learning

Active Learning algorithms using GPCalculator. The GPCalculator takes as input parameter some training Atoms — i.e. a list of Atoms which would contain previously calculated energies and forces. This calculator can be attached to an Atoms object using atoms.set_calculator(calc) similarly to the other ASE calculators. The user can ask for the predicted energies, forces and uncertainties using atoms.get_potential_energy(), atoms.get_forces() and atoms.get_calculator().results[‘uncertainty’]. The idea is to make a simple regression core for the ASE community, in which training and testing would be simple and will mostly use ASE Atoms objects as inputs. Here, we also include to different active learning recipes using the GPCalculator for accelerating: (1) structural minimization and (2) NEBs.