def read_nonstd_ext_xyz(f):
n_atoms = None
R, z, E, F = [], [], [], []
for i, line in enumerate(f):
line = line.strip()
if not n_atoms:
n_atoms = int(line)
print('Number atoms per geometry: {:,}'.format(n_atoms))
file_i, line_i = divmod(i, n_atoms + 2)
if line_i == 1:
try:
e = float(line)
except ValueError:
pass
else:
E.append(e)
cols = line.split()
if line_i >= 2:
R.append(list(map(float, cols[1:4])))
if file_i == 0: # first molecule
z.append(io._z_str_to_z_dict[cols[0]])
F.append(list(map(float, cols[4:7])))
if file_i % 1000 == 0:
sys.stdout.write('\rNumber geometries found so far: {:,}'.format(file_i))
sys.stdout.flush()
sys.stdout.write('\rNumber geometries found so far: {:,}'.format(file_i))
sys.stdout.flush()
print()
R = np.array(R).reshape(-1, n_atoms, 3)
z = np.array(z)
E = None if not E else np.array(E)
F = np.array(F).reshape(-1, n_atoms, 3)
if F.shape[0] != R.shape[0]:
sys.exit(
ui.color_str('[FAIL]', fore_color=ui.RED, bold=True)
+ ' Force labels are missing from dataset or are incomplete!'
)
f.close()
return (R, z, E, F)
'--overwrite',
dest='overwrite',
action='store_true',
help='overwrite existing xyz dataset file',
)
args = parser.parse_args()
dataset_path, dataset = args.dataset
name = os.path.splitext(os.path.basename(dataset_path))[0]
dataset_file_name = name + '.xyz'
xyz_exists = os.path.isfile(dataset_file_name)
if xyz_exists and args.overwrite:
print(
ui.color_str('[INFO]', bold=True) +
' Overwriting existing xyz dataset file.')
if not xyz_exists or args.overwrite:
print(
ui.color_str('[INFO]', bold=True) +
' Writing dataset to \'{}\'...'.format(dataset_file_name))
else:
sys.exit(
ui.color_str('[FAIL]', fore_color=ui.RED, bold=True) +
' Dataset \'{}\' already exists.'.format(dataset_file_name))
R = dataset['R']
z = dataset['z']
F = dataset['F']
lattice = dataset['lattice'] if 'lattice' in dataset else None
def sgdml_all_default(train_indices, args):
from sgdml.cli import create, train, validate, select, test
from sgdml.utils import ui, io
ui.print_step_title("STEP 1", "Cross-validation task creation")
task_dir = create(**args)
dataset = args["dataset"][1]
if (train_indices is not None) and not (type(train_indices) == int):
# CHANGE TRAINING INDICES
# AND RELATED ARRAYS
R_train = dataset["R"][train_indices]
F_train = dataset["F"][train_indices]
E_train = dataset["E"][train_indices]
for file in os.listdir(task_dir):
if file.endswith(".npz"):
name = os.path.join(task_dir, file)
a = dict(np.load(name, allow_pickle=True))
a["R_train"] = R_train
a["F_train"] = F_train
if "E_train" in a:
a["E_train"] = E_train
a["idxs_train"] = train_indices
np.savez_compressed(name, **a)
ui.print_step_title("STEP 2", "Training")
task_dir_arg = io.is_dir_with_file_type(task_dir, "task")
args["task_dir"] = task_dir_arg
model_dir_or_file_path = train(**args)
ui.print_step_title("STEP 3", "Validation")
model_dir_arg = io.is_dir_with_file_type(model_dir_or_file_path,
"model",
or_file=True)
valid_dataset = args["valid_dataset"]
validate(
model_dir_arg,
valid_dataset,
overwrite=False,
max_processes=args["max_processes"],
use_torch=args["use_torch"],
)
ui.print_step_title("STEP 4", "Hyper-parameter selection")
model_file_name = select(model_dir_arg, args["overwrite"],
args["max_processes"], args["model_file"])
ui.print_step_title("STEP 5", "Testing")
model_dir_arg = io.is_dir_with_file_type(model_file_name,
"model",
or_file=True)
test_dataset = args["test_dataset"]
test(
model_dir_arg,
test_dataset,
args["n_test"],
overwrite=False,
max_processes=args["max_processes"],
use_torch=args["use_torch"],
)
print("\n" + ui.color_str(
" DONE ", fore_color=ui.BLACK, back_color=ui.GREEN, bold=True) +
" Training assistant finished sucessfully.")
print(" This is your model file: '{}'".format(model_file_name))
if "glob" in globals():
global glob
del glob
'-o',
'--overwrite',
dest='overwrite',
action='store_true',
help='overwrite existing dataset file',
)
args = parser.parse_args()
dataset = args.dataset
name = os.path.splitext(os.path.basename(dataset.name))[0]
dataset_file_name = name + '.npz'
dataset_exists = os.path.isfile(dataset_file_name)
if dataset_exists and args.overwrite:
print(
ui.color_str('[INFO]', bold=True) +
' Overwriting existing dataset file.')
if not dataset_exists or args.overwrite:
print('Writing dataset to \'{}\'...'.format(dataset_file_name))
else:
sys.exit(
ui.color_str('[FAIL]', fore_color=ui.RED, bold=True) +
' Dataset \'{}\' already exists.'.format(dataset_file_name))
mols = read(dataset.name, index=':')
lattice, R, z, E, F = None, None, None, None, None
calc = mols[0].get_calculator()
print("\rNumber geometries found: {:,}\n".format(len(mols)))
'-o',
'--overwrite',
dest='overwrite',
action='store_true',
help='overwrite existing dataset file',
)
args = parser.parse_args()
dataset = args.dataset
name = os.path.splitext(os.path.basename(dataset.name))[0]
dataset_file_name = name + '.npz'
dataset_exists = os.path.isfile(dataset_file_name)
if dataset_exists and args.overwrite:
print(ui.color_str('[INFO]', bold=True) + ' Overwriting existing dataset file.')
if not dataset_exists or args.overwrite:
print('Writing dataset to \'{}\'...'.format(dataset_file_name))
else:
sys.exit(
ui.color_str('[FAIL]', fore_color=ui.RED, bold=True)
+ ' Dataset \'{}\' already exists.'.format(dataset_file_name)
)
lattice, R, z, E, F = None, None, None, None, None
mols = read(dataset.name, index=':')
calc = mols[0].get_calculator()
is_extxyz = calc is not None
if is_extxyz:
def sgdml_all_default(train_indices, args):
from sgdml.cli import create, train, validate, select, test
from sgdml.utils import ui, io
ui.print_step_title('STEP 1', 'Cross-validation task creation')
task_dir = create(**args)
dataset = args['dataset'][1]
if (train_indices is not None) and not (type(train_indices) == int):
# CHANGE TRAINING INDICES
# AND RELATED ARRAYS
R_train = dataset['R'][train_indices]
F_train = dataset['F'][train_indices]
E_train = dataset['E'][train_indices]
for file in os.listdir(task_dir):
if file.endswith('.npz'):
name = os.path.join(task_dir, file)
a = dict(np.load(name, allow_pickle=True))
a['R_train'] = R_train
a['F_train'] = F_train
if 'E_train' in a:
a['E_train'] = R_train
a['idxs_train'] = train_indices
np.savez_compressed(name, **a)
ui.print_step_title('STEP 2', 'Training')
task_dir_arg = io.is_dir_with_file_type(task_dir, 'task')
args['task_dir'] = task_dir_arg
model_dir_or_file_path = train(**args)
ui.print_step_title('STEP 3', 'Validation')
model_dir_arg = io.is_dir_with_file_type(model_dir_or_file_path,
'model',
or_file=True)
valid_dataset = args['valid_dataset']
validate(
model_dir_arg,
valid_dataset,
overwrite=False,
max_processes=args['max_processes'],
use_torch=args['use_torch'],
)
ui.print_step_title('STEP 4', 'Hyper-parameter selection')
model_file_name = select(model_dir_arg, args['overwrite'],
args['max_processes'], args['model_file'])
ui.print_step_title('STEP 5', 'Testing')
model_dir_arg = io.is_dir_with_file_type(model_file_name,
'model',
or_file=True)
test_dataset = args['test_dataset']
test(
model_dir_arg,
test_dataset,
args['n_test'],
overwrite=False,
max_processes=args['max_processes'],
use_torch=args['use_torch'],
)
print('\n' + ui.color_str(
' DONE ', fore_color=ui.BLACK, back_color=ui.GREEN, bold=True) +
' Training assistant finished sucessfully.')
print(' This is your model file: \'{}\''.format(model_file_name))
if "glob" in globals():
global glob
del glob
