def testEntry(self):
entries = []
for i, f in enumerate(self.iter_path):
vi = VaspInput.from_directory(f)
ls = LabeledSystem(os.path.join(f, 'OUTCAR'))
attrib = loadfn(os.path.join(f, 'job.json'))
comp = vi['POSCAR'].structure.composition
entry = Entry(comp,
'vasp',
vi.as_dict(),
ls.as_dict(),
entry_id='pku-' + str(i),
attribute=attrib)
entries.append(entry)
self.assertEqual(len(entries), len(self.ref_entries))
ret0 = entries[0]
r0 = self.ref_entries[0]
self.assertEqual(Incar.from_dict(ret0.inputs['INCAR']),
Incar.from_dict(r0.inputs['INCAR']))
self.assertEqual(str(r0.inputs['KPOINTS']),
str(Kpoints.from_dict(ret0.inputs['KPOINTS'])))
self.assertEqual(ret0.inputs['POTCAR'], r0.inputs['POTCAR'].as_dict())
self.assertEqual(
Poscar.from_dict(ret0.inputs['POSCAR']).structure,
r0.inputs['POSCAR'].structure)
self.assertEqual(ret0.entry_id, 'pku-0')
def compute(self, output_dir):
outcar = os.path.join(output_dir, 'OUTCAR')
if not os.path.isfile(outcar):
dlog.warning("cannot find OUTCAR in " + output_dir + " skip")
return None
else:
ls = LabeledSystem(outcar)
stress = []
with open(outcar, 'r') as fin:
lines = fin.read().split('\n')
for line in lines:
if 'in kB' in line:
stress_xx = float(line.split()[2])
stress_yy = float(line.split()[3])
stress_zz = float(line.split()[4])
stress_xy = float(line.split()[5])
stress_yz = float(line.split()[6])
stress_zx = float(line.split()[7])
stress.append([])
stress[-1].append([stress_xx, stress_xy, stress_zx])
stress[-1].append([stress_xy, stress_yy, stress_yz])
stress[-1].append([stress_zx, stress_yz, stress_zz])
outcar_dict = ls.as_dict()
outcar_dict['data']['stress'] = {
"@module": "numpy",
"@class": "array",
"dtype": "float64",
"data": stress
}
return outcar_dict
def build_deepmd(path,nsw):
ls = LabeledSystem(os.path.join(path, 'outcar'),fmt='outcar')
deepmd = os.path.join(path,'deepmd')
if nsw <= 4: # we know nsw must > 100
set_size = 1
print("{0} has only {1}".format(path,nsw))
if nsw > 4:
set_size = nsw//4 # 25% used as , but if say 82, then 20, 20, 20, 2, too less
ls.to_deepmd_npy(deepmd,set_size=set_size)
def build_deepmd(path,nsw):
ls=LabeledSystem(os.path.join(path, 'OUTCAR'),fmt='outcar')
deepmd = os.path.join(path,'deepmd')
if nsw <= 2000: # we know nsw must > 100
set_size = nsw//2
if nsw > 2000:
set_size = 1000
ls.to_deepmd_npy(deepmd,set_size=set_size)
if nsw>3000:
check_sets(deepmd)
def load(cls, filename, Cls=None):
with open(filename, 'r') as f:
fc = f.read()
jc = json.loads(fc)
composition = jc['composition']
calculator = jc['calculator']
if calculator.lower() == 'vasp':
try:
inputs = VaspInput.from_dict(jc['inputs']).as_dict()
except:
inputs = jc['inputs']
warnings.warn("""Inproperly configure of POTCAR !
Returned instance cannot be used
as input for from_dict() method """)
else:
if Cls:
inputs = Cls.from_dict(jc['inputs']).as_dict()
else:
raise RuntimeError("inputs decoder must be given")
data = LabeledSystem.from_dict(jc['data']).as_dict()
attribute = jc['attribute']
entry_id = jc['entry_id']
tag = jc['tag']
return cls(composition, calculator, inputs, data, entry_id, attribute,
tag)
def to_system(self, data, **kwargs):
"""
convert system to list, usefull for data collection
"""
from dpdata import System, LabeledSystem
if 'forces' in data:
system = LabeledSystem(data=data)
else:
system = System(data=data)
if len(system) == 0:
return []
if len(system) == 1:
return [system]
else:
systems = []
for ii in range(len(system)):
systems.append(system.sub_system([ii]))
return systems
def test():
from monty.serialization import dumpfn, loadfn
from monty.json import MontyDecoder, MontyEncoder
from pymatgen.io.vasp.inputs import PotcarSingle, Potcar
vi = VaspInput.from_directory('.')
ls = LabeledSystem('OUTCAR', fmt='vasp/outcar')
en0 = Entry('Al',
'vasp',
inputs=vi.as_dict(),
data=ls.as_dict(),
entry_id='pku-1')
print(en0)
fname = 'pku-1.json'
dumpfn(en0.as_dict(), fname, indent=4)
en1 = Entry.load(fname)
#vin=VaspInput.from_dict(en1.inputs)
#vin.write_input('./new')
print(en1)
print(en1.as_dict())
def _parsing_vasp(paths, id_prefix, iters=True):
entries = []
icount = 0
for path in paths:
f_outcar = os.path.join(path, 'OUTCAR')
f_job = os.path.join(path, 'job.json')
try:
vi = VaspInput.from_directory(path)
if os.path.isfile(f_job):
attrib = loadfn(f_job)
else:
attrib = {}
if iters and attrib:
tmp_ = path.split('/')[-1]
iter_info = tmp_.split('.')[1]
task_info = tmp_.split('.')[-1]
attrib['iter_info'] = iter_info
attrib['task_info'] = task_info
else:
pass
comp = vi['POSCAR'].structure.composition
ls = LabeledSystem(f_outcar)
lss = ls.to_list()
for ls in lss:
if id_prefix:
eid = id_prefix + "_" + str(icount)
else:
eid = str(uuid4())
entry = Entry(comp,
'vasp',
vi.as_dict(),
ls.as_dict(),
attribute=attrib,
entry_id=eid)
entries.append(entry)
icount += 1
except:
dlog.info("failed here : %s" % path)
return entries
def extract_outcar(outcar):
"""
extract e, f, v
"""
### get confgis that were recalculated
ls = LabeledSystem(outcar, fmt='outcar')
fp = open(outcar)
fp.readline()
nsw_sel = fp.readline()
if 'nsw_sel' in nsw_sel:
print('file generated by merge_out.py')
tmp = nsw_sel.split('=')[1].strip().split(' ')
nsw_sel = [int(tmp_idx) for tmp_idx in tmp]
### get confgis that were recalculated
etot = ls['energies']
nsw = np.array(nsw_sel).astype(int) - 1 # relative nsw, starting from 0
stress = ls['virials']
forces = ls['forces']
return etot, stress, forces, nsw
def build_deepmd(path, nsw, outcar, deepmd):
ls = LabeledSystem(outcar, fmt='outcar')
"""
sub_ls = ls.sub_system(idx)
"""
if args.idx:
print("index file provided")
idx = np.loadtxt(args.idx).astype(int)
# ls = ls.sub_system(idx)
if args.vaspidx:
print("vasp index file provided")
vaspidx = np.loadtxt(args.vaspidx)
fp = open(outcar)
fp.readline()
nsw_sel = fp.readline()
if 'nsw_sel' in nsw_sel:
print('file generated by merge_out.py')
# print(nsw_sel)
tmp = nsw_sel.split('=')[1].strip().split(' ')
nsw_sel = [int(tmp_idx) for tmp_idx in tmp]
idx = []
for i in range(len(nsw_sel)):
if nsw_sel[i] in vaspidx:
idx.append(i)
else:
print('OUTCAR file generated by VASP')
idx = vaspidx - 1
idx2 = [i for i in range(len(ls)) if i not in idx]
ls2 = ls.sub_system(idx2)
ls = ls.sub_system(idx)
deepmd = os.path.join(path, deepmd)
if args.batchsize:
set_size = args.batchsize
else:
if nsw <= 4: # we know nsw must > 100
set_size = 1
print("{0} has only {1}".format(path, nsw))
if nsw > 4:
set_size, _ = best_size(
nsw
) # 25% used as , but if say 82, then 20, 20, 20, 2, too less
ls.to_deepmd_npy(deepmd, set_size=set_size)
if args.test:
ls2.to_deepmd_npy('test_tmp', set_size=100000)
shutil.copytree('test_tmp/set.000', os.path.join(deepmd, 'set.001'))
shutil.rmtree('test_tmp')
from glob import glob
from tqdm import tqdm
"""
process multi systems
"""
fs = glob('iter.0000[4-7]*/02.fp/task*/OUTCAR')
maxf = 3.0
ms = MultiSystems()
ic = 0
vacuum_size = 13
for f in tqdm(fs):
if check_cluster(f.replace('OUTCAR', 'POSCAR'), vacuum_size, fmt='POSCAR'):
print(f)
continue
try:
ls = LabeledSystem(f)
except:
print(f)
continue
if len(ls) > 0:
if ls.sub_system([0]).data['forces'].max() > maxf:
pass
else:
ic += 1
ms.append(ls)
print(len(fs))
print(ic)
ms.to_deepmd_raw('deepmd-f%s' % maxf)
ms.to_deepmd_npy('deepmd-f%s' % maxf)
def build_deepmd_frames(path, outcar, deepmd):
"""
sub_ls = ls.sub_system(idx)
"""
try:
ls = LabeledSystem(outcar, fmt=args.format)
except:
ls = System(outcar, fmt=args.format)
if args.exclude:
oldsize = len(ls)
idx_new = [i for i in range(len(ls)) if i not in args.exclude]
ls = ls.sub_system(idx_new)
newsize = len(ls)
print('{0}/{1} is selected'.format(newsize, oldsize))
if args.force_limit:
fmin = min(args.force_limit)
fmax = max(args.force_limit)
print("force limit imposed, force in between {0}, {1}".format(
fmin, fmax))
idx_new = []
exclude = []
for i in range(len(ls)):
forces = ls[i].data['forces']
if forces.min() >= fmin and forces.max() <= fmax:
idx_new.append(i)
else:
exclude.append(i)
print('excluded frames', exclude)
print('{0} / {1} is selected'.format(len(idx_new), len(ls)))
ls = ls.sub_system(idx_new)
if args.idx:
print("index file provided")
idx = np.loadtxt(args.idx).astype(int)
elif (not args.idx) and args.vaspidx:
print("vasp index file provided")
vaspidx = np.loadtxt(args.vaspidx)
fp = open(outcar)
fp.readline()
nsw_sel = fp.readline()
if 'nsw_sel' in nsw_sel:
print('file generated by merge_out.py')
tmp = nsw_sel.split('=')[1].strip().split(' ')
nsw_sel = [int(tmp_idx) for tmp_idx in tmp]
idx = []
for i in range(len(nsw_sel)):
if nsw_sel[i] in vaspidx:
idx.append(i)
else:
print('OUTCAR file generated by VASP')
idx = vaspidx - 1
else:
print("split train and test by ratio {0} : {1}".format(
args.train_test_ratio, 1))
train_size = round(
len(ls) * (args.train_test_ratio) / (args.train_test_ratio + 1))
idx = np.random.choice(range(len(ls)), train_size, replace=False)
idx.sort()
idx2 = [i for i in range(len(ls)) if i not in idx] # test
ls2 = ls.sub_system(idx2) # test
ls = ls.sub_system(idx)
deepmd = os.path.join(path, deepmd)
ls.to_deepmd_npy(deepmd,
set_size=1000000) # give a *large* value, default is 5000
if len(ls2) == 0:
print('test set has no data')
elif args.savetest and len(ls2) > 0:
ls2.to_deepmd_npy('test_tmp', set_size=1000000)
shutil.copytree('test_tmp/set.000', os.path.join(deepmd, 'set.001'))
shutil.rmtree('test_tmp')
print(force1.min(axis=1))
print(min(force1.min(axis=1)), max(force1.max(axis=1)))
print(np.argmin(force1.min(axis=1)), np.argmax(force1.max(axis=1)))
#force=np.load('/Users/jiedeng/GD/papers/pv3_crystallization/post_nn/generate_new_poscar/pairs/untitled folder/sisi/deepmd/set.000/force.npy')
#max_f = []
#for f in force:
# max_f.append(np.max(np.max(f)))
#print(max_f)
#plt.plot(max_f)
tmp = '/Users/jiedeng/GD/papers/pv3_crystallization/post_nn/exsolution_pert/u.project.ESS.lstixrud.jd848.pv+hf.dp-train.lmp_run.6k.rp5.160-cpu.pert.10k_good_p3.recal/deepmd_all'
tmp2 = '/Users/jiedeng/GD/papers/pv3_crystallization/post_nn/exsolution_pert/u.home.j.jd848.project-lstixrud.metad.3rd.recal/deepmd/'
ls = LabeledSystem(tmp2,fmt='deepmd/npy')
from dpdata import LabeledSystem
import numpy as np
ls = LabeledSystem('deepmd_ttr2',fmt='deepmd/npy')
idx = list(range(len(ls)))
idx.remove(14)
ls2 = ls.sub_system(idx)
ls2.to_deepmd_npy('test')
fparam = np.load('deepmd_ttr2/set.001/fparam.npy')
np.save('test/set.000/fparam.npy',fparam[idx])
cp -r deepmd_ttr2 deepmd_ttr2_tmp
rm -r deepmd_ttr2_tmp/set.000
from dpdata import LabeledSystem
import sys
sys.path.append(
'/Users/jiedeng/Documents/ml/deepmd-kit/my_example/codes/Data.py')
from Data import DataSets
from sklearn.preprocessing import StandardScaler
from sklearn.decomposition import PCA
import dpdata
vasp_multi_systems = dpdata.MultiSystems.from_dir(
dir_name=
'/Users/jiedeng/Documents/tmp/jd848/project_folder/pv+hf/3k/solid1/r3-3k/',
file_name='OUTCAR',
fmt='vasp/outcar')
from dpdata import LabeledSystem
ls = LabeledSystem(
'/Users/jiedeng/Documents/tmp/jd848/project_folder/pv+hf/3k/solid1/r3-3k/OUTCAR',
fmt='outcar')
print(ls.data['coords'].shape) # (5000, 160, 3)
scaler = StandardScaler()
"""
we get n sample, we need find the most important or representative coord so that
them can represent the rest of features.
1) z-score variables
2) eigendecomposition of covariance matrix, covriance matrix should be n*n, not 480*480
3) sort eigenvalues
3) projection of the original normalized data onto the PCA space
If we use above protocal, dat should be dat
But if we use PCA module directly, input should be dat.T
"""
### benchmark
from dpdata import LabeledSystem,MultiSystems
from glob import glob
"""
process multi systems
"""
try:
ls=LabeledSystem('OUTCAR')
except:
print(f)
ms.to_deepmd_raw('deepmd')
ms.to_deepmd_npy('deepmd')
from dpdata import System, LabeledSystem, MultiSystems
import os
fp = open('folders_to_merge', 'r')
folders_org = fp.readlines()
folders = []
fp.close()
for i in range(len(folders_org)):
if '#' in folders_org[i] or folders_org[i] == '\n':
pass
else:
folders.append(folders_org[i].replace('\n', ''))
for path in folders:
pwd = os.getcwd()
os.chdir(path)
print("process ", path)
#s=System('POSCAR',fmt='poscar')
ls = LabeledSystem('OUTCAR', fmt='outcar')
ls.to_deepmd_raw('deepmd')
ls.to_deepmd_npy('deepmd', set_size=1000)
os.chdir(pwd)
print("done ", path)
print("done")
def _parsing_vasp(paths, config_info_dict, id_prefix, iters=True):
entries = []
icount = 0
if iters:
iter_record = []
iter_record_new = []
try:
with open("record.database", "r") as f_record:
iter_record = [i.split()[0] for i in f_record.readlines()]
iter_record.sort()
dlog.info("iter_record")
dlog.info(iter_record)
except:
pass
for path in paths:
try:
f_outcar = os.path.join(path, 'OUTCAR')
f_job = os.path.join(path, 'job.json')
tmp_iter = path.split('/')[-3]
if (tmp_iter in iter_record) and (tmp_iter != iter_record[-1]):
continue
if tmp_iter not in iter_record_new:
iter_record_new.append(tmp_iter)
vi = VaspInput.from_directory(path)
if os.path.isfile(f_job):
attrib = loadfn(f_job)
else:
attrib = {}
if iters and attrib:
# generator/Cu/iter.000031/02.fp/task.007.000000
tmp_ = path.split('/')[-1]
#config_info=tmp_.split('.')[1]
task_info = tmp_.split('.')[-1]
tmp_iter = path.split('/')[-3]
iter_info = tmp_iter.split('.')[-1]
sys_info = path.split('/')[-4]
config_info_int = int(tmp_.split('.')[1])
for (key, value) in config_info_dict.items():
if config_info_int in value:
config_info = key
attrib['config_info'] = config_info
attrib['task_info'] = task_info
attrib['iter_info'] = iter_info
attrib['sys_info'] = sys_info
with open(f_outcar, "r") as fin_outcar:
infile_outcar = fin_outcar.readlines()
for line in infile_outcar:
if "running on" in line:
attrib["core"] = int(line.split()[2])
if "Elapse" in line:
attrib["wall_time"] = float(line.split()[-1])
if "executed on" in line:
attrib["date"] = line.split()[-2]
attrib["clocktime"] = line.split()[-1]
dlog.info("Attrib")
dlog.info(attrib)
comp = vi['POSCAR'].structure.composition
ls = LabeledSystem(f_outcar)
lss = ls.to_list()
for ls in lss:
if id_prefix:
eid = id_prefix + "_" + str(icount)
else:
eid = str(uuid4())
entry = Entry(comp,
'vasp',
vi.as_dict(),
ls.as_dict(),
attribute=attrib,
entry_id=eid)
entries.append(entry)
icount += 1
except Exception:
#dlog.info(str(Exception))
dlog.info("failed for %s" % (path))
#pass
if iters:
iter_record.sort()
iter_record_new.sort()
with open("record.database", "w") as fw:
for line in iter_record:
fw.write(line + "\n")
for line in iter_record_new:
fw.write(line + "\n")
return entries
from dpdata import System, LabeledSystem, MultiSystems
#s = System('POSCAR', fmt='poscar')
#print(s)
ls = LabeledSystem('OUTCAR', fmt='outcar')
"""
if len(ls)%2==0:
size = int(len(ls)/2)
else:
size = int(len(ls)/2) + 1
"""
ls.to_deepmd_raw('.')
#ls.to_deepmd_npy('deepmd', set_size=size)
from dpdata import LabeledSystem, MultiSystems
from glob import glob
"""
process multi systems
"""
fs = glob('./*/[0-9]*/OUTCAR')
ms = MultiSystems()
for f in fs:
try:
ls = LabeledSystem(f)
except:
print(f)
if len(ls) > 0:
ms.append(ls)
ms.to_deepmd_raw('deepmd')
ms.to_deepmd_npy('deepmd')
from dpdata import LabeledSystem, MultiSystems
from glob import glob
from tqdm import tqdm
"""
process multi systems
"""
fs = glob('iter.0000[3-5]*/02.fp/task*/OUTCAR')
maxf = 1.0
ms = MultiSystems()
ic = 0
for f in tqdm(fs):
try:
ls = LabeledSystem(f)
except:
print(f)
if len(ls) > 0:
st = ls.to_pymatgen_structure()[0]
z = st.cart_coords[:, 2]
if st.lattice.c - (z.max() - z.min()) < 14:
pass
else:
if ls.sub_system([0]).data['forces'].max() > maxf:
pass
else:
ic += 1
ms.append(ls)
print(len(fs))
print(ic)
ms.to_deepmd_raw('deepmd-f%s' % maxf)
ms.to_deepmd_npy('deepmd-f%s' % maxf)
from glob import glob
from dpdata import LabeledSystem
from monty.serialization import dumpfn, loadfn
from tqdm import tqdm
fs = glob('usefull-[1-3]/sys-*/OUTCAR')
entries = []
for f in tqdm(fs):
ls = LabeledSystem(f)
ls.sub_system([-1]).to_pymatgen_ComputedStructureEntry()
entry = ls.sub_system([-1]).to_pymatgen_ComputedStructureEntry()[0]
entries.append(entry)
dumpfn(entries, 'all-vasp-entries.json')
