def mgp_model(gp_model):
"""
test the init function
"""
grid_params = {}
if 'twobody' in gp_model.kernels:
grid_params['twobody'] = {'grid_num': [64], 'lower_bound': [0.1]}
if 'threebody' in gp_model.kernels:
grid_params['threebody'] = {
'grid_num': [16] * 3,
'lower_bound': [0.1] * 3
}
species_list = [1, 2, 3]
lammps_location = f'tmp_lmp'
mapped_model = MappedGaussianProcess(grid_params=grid_params,
unique_species=species_list,
n_cpus=1,
lmp_file_name=lammps_location,
var_map='simple')
# import flare.mgp.mapxb
# flare.mgp.mapxb.global_use_grid_kern = False
mapped_model.build_map(gp_model)
yield mapped_model
del mapped_model
def test_mgp_gpfa(all_mgp, all_gp):
"""
Ensure that passing in an MGP also works for the trajectory trainer
:param all_mgp:
:param all_gp:
:return:
"""
np.random.seed(10)
gp_model = get_gp("3", "mc", False)
gp_model.set_L_alpha()
grid_num_3 = 3
lower_cut = 0.01
grid_params_3b = {
"lower_bound": [lower_cut] * 3,
"grid_num": [grid_num_3] * 3,
"svd_rank": "auto",
}
grid_params = {"load_grid": None, "update": False}
grid_params["threebody"] = grid_params_3b
unique_species = gp_model.training_statistics["species"]
mgp_model = MappedGaussianProcess(grid_params=grid_params,
unique_species=unique_species,
n_cpus=1)
mgp_model.build_map(gp_model)
nenv = 10
cell = np.eye(3)
struc, f = get_random_structure(cell, unique_species, nenv)
struc.forces = np.array(f)
frames = [struc]
tt = TrajectoryTrainer(
frames,
mgp_model,
rel_std_tolerance=0,
abs_std_tolerance=0,
abs_force_tolerance=1e-8,
print_training_plan=True,
)
assert tt.gp_is_mapped is True
tt.run()
# Test that training plan is properly written
with open("gp_from_aimd_training_plan.json", "r") as f:
plan = json.loads(f.readline())
assert isinstance(plan["0"], list)
assert len(plan["0"]) == len(struc)
assert [p[0] for p in plan["0"]] == list(range(len(struc)))
for f in glob(f"gp_from_aimd*"):
remove(f)