Generating samples

Generating samples with random noise

To create samples to train our model, we need first to generate structures

from carcara.sampler.random_displacements import RandomDisplacements
from ase.build import bulk
from ase.calculator.emt import EMT

atoms = bulk("Au", "fcc", a=4.08, cubic=True).repeat([2,2,2])
calc = EMT()

generator = RandomDisplacements(atoms=atoms, calculator=calculator, seed=42)
generator.relax_structure(fmax=0.01, relax_cell=True)
generator.generate_samples(num_samples=100,
                           noise_type='normal',
                           noise_level_pos=0.2,
                           noise_level_cell=0.2,
                           cell_mode='all',
                           compute_energy_and_forces=True)
generator.write_xyz("noise_samples.xyz")

../_images/noisy.gif — Figure 1: Sample generation using uniform random displacements (0.4 Å) for atomic positions and lattice components.

Calculating energies and forces with MACE

Analyzing the diversity of the samples generated