def dump(self, dump_filename, mkdir=True):
"""Write object to binary file using cPickle. Read back
with :func:`~pwtools.io.read_pickle`."""
# Dumping with protocol "2" is supposed to be the fastest binary format
# writing method. Probably, this is platform-specific.
if mkdir:
dr = os.path.dirname(dump_filename)
if dr != '':
common.makedirs(dr)
cPickle.dump(self, open(dump_filename, 'wb'), protocol=2)
def dump(self, dump_filename, mkdir=True):
"""Write object to binary file using pickle. Read back
with :func:`~pwtools.io.read_pickle`."""
# Dumping with protocol "2" is supposed to be the fastest binary format
# writing method. Probably, this is platform-specific.
if mkdir:
dr = os.path.dirname(dump_filename)
if dr != '':
common.makedirs(dr)
pickle.dump(self, open(dump_filename, 'wb'), protocol=2)
def test_lammps_calculator():
if not have_ase():
skip("no ASE found, skipping test")
elif not have_lmp():
skip("no lammps found, skipping test")
else:
at = get_atoms_with_calc_lammps()
at.rattle(stdev=0.001, seed=int(time.time()))
common.makedirs(at.calc.directory)
print(common.backtick("cp -v utils/lammps/AlN.tersoff {p}/".format(
p=at.calc.directory)))
print("scf")
forces = at.get_forces()
etot = at.get_potential_energy()
stress = at.get_stress(voigt=False) # 3x3
st = io.read_lammps_md_txt(at.calc.label + '.out')[0]
assert np.allclose(forces, st.forces)
assert np.allclose(etot, st.etot)
assert np.allclose(st.stress, -stress * constants.eV_by_Ang3_to_GPa,
atol=1e-10)
print("relax")
from ase.optimize import BFGS
opt = BFGS(at, maxstep=0.04)
opt.run(fmax=0.001, steps=10)
coords_frac = parse.arr2d_from_txt("""
3.3333341909920072e-01 6.6666683819841532e-01 4.4325467247779138e-03
6.6666681184103216e-01 3.3333362368205072e-01 5.0443254824788963e-01
3.3333341909918301e-01 6.6666683819838046e-01 3.8356759709402671e-01
6.6666681184101539e-01 3.3333362368201563e-01 8.8356759861713752e-01
""")
assert np.allclose(coords_frac, at.get_scaled_positions(), atol=1e-2)
# at least 1 backup files must exist
assert os.path.exists(at.calc.infile + '.0')
assert os.path.exists(at.calc.outfile + '.0')
assert os.path.exists(at.calc.dumpfile + '.0')
assert os.path.exists(at.calc.structfile + '.0')
def test_lammps_calculator():
if not have_ase():
skip("no ASE found, skipping test")
elif not have_lmp():
skip("no lammps found, skipping test")
else:
at = get_atoms_with_calc_lammps()
at.rattle(stdev=0.001, seed=int(time.time()))
common.makedirs(at.calc.directory)
print common.backtick("cp -v utils/lammps/AlN.tersoff {p}/".format(
p=at.calc.directory))
print "scf"
forces = at.get_forces()
etot = at.get_potential_energy()
stress = at.get_stress(voigt=False) # 3x3
st = io.read_lammps_md_txt(at.calc.label + '.out')[0]
assert np.allclose(forces, st.forces)
assert np.allclose(etot, st.etot)
assert np.allclose(st.stress, -stress * constants.eV_by_Ang3_to_GPa,
atol=1e-10)
print "relax"
from ase.optimize import BFGS
opt = BFGS(at, maxstep=0.04)
opt.run(fmax=0.001, steps=10)
coords_frac = parse.arr2d_from_txt("""
3.3333341909920072e-01 6.6666683819841532e-01 4.4325467247779138e-03
6.6666681184103216e-01 3.3333362368205072e-01 5.0443254824788963e-01
3.3333341909918301e-01 6.6666683819838046e-01 3.8356759709402671e-01
6.6666681184101539e-01 3.3333362368201563e-01 8.8356759861713752e-01
""")
assert np.allclose(coords_frac, at.get_scaled_positions(), atol=1e-2)
# at least 1 backup files must exist
assert os.path.exists(at.calc.infile + '.0')
assert os.path.exists(at.calc.outfile + '.0')
assert os.path.exists(at.calc.dumpfile + '.0')
assert os.path.exists(at.calc.structfile + '.0')
a, c = axes_flat[ii, :]
fc = 550 - 50 * V[ii] / Vmax
phdos.append(np.array([freq, gauss(freq - fc, 100) * 0.01]).T)
gibbs = Gibbs(T=T,
P=P,
etot=etot,
phdos=phdos,
axes_flat=axes_flat,
volfunc_ax=volfunc_ax,
case=case,
dosarea=None)
gibbs.set_fitfunc('C',
lambda x, y: num.Spline(x, y, s=None, k=5, eps=1e-5))
g = gibbs.calc_G(calc_all=True)
common.makedirs('../files/gibbs/2d')
io.write_h5('../files/gibbs/2d/%s.h5' % gethostname(),
filt_dct(g),
mode='w')
# 1d case
case = '1d'
V = np.linspace(10, 20, nax)
axes_flat = V**(1 / 3.) # cubic
volfunc_ax = lambda x: x[0]**3.0
etot = (V - V.mean())**2
fcenter = 450 + 100 * (axes_flat - axes_flat.min())
# fake phonon dos data (Gaussian), shift to lower freq for higher volume
phdos = [np.array([freq, gauss(freq - fc, 100)]).T for fc in fcenter[::-1]]
gibbs = Gibbs(T=T,
def write_input(self, mode='a', backup=True, sleep=0, excl=True):
"""
Create calculation dir(s) for each parameter set and write input files
based on ``templates``. Write sqlite database storing all relevant
parameters. Write (bash) shell script to start all calculations (run
locally or submitt batch job file, depending on ``machine.subcmd``).
Parameters
----------
mode : str, optional
Fine tune how to write input files (based on ``templates``) to calc
dirs calc_foo/0/, calc_foo/1/, ... . Note that this doesn't change
the base dir calc_foo at all, only the subdirs for each calc.
{'a', 'w'}
| 'a': Append mode (default). If a previous database is found, then
| subsequent calculations are numbered based on the last 'idx'.
| calc_foo/0 # old
| calc_foo/1 # old
| calc_foo/2 # new
| calc_foo/3 # new
| 'w': Write mode. The target dirs are purged and overwritten. Also,
| the database (self.dbfn) is overwritten. Use this to
| iteratively tune your inputs, NOT for working on already
| present results!
| calc_foo/0 # new
| calc_foo/1 # new
backup : bool, optional
Before writing anything, do a backup of self.calc_dir if it already
exists.
sleep : int, optional
For the script to start (submitt) all jobs: time in seconds for the
shell sleep(1) commmand.
excl : bool
If in append mode, a file <calc_root>/excl_push with all indices of
calculations from old revisions is written. Can be used with
``rsync --exclude-from=excl_push`` when pushing appended new
calculations to a cluster.
"""
assert mode in ['a', 'w'], "Unknown mode: '%s'" %mode
if os.path.exists(self.dbfn):
if backup:
common.backup(self.dbfn)
if mode == 'w':
os.remove(self.dbfn)
have_new_db = not os.path.exists(self.dbfn)
common.makedirs(self.calc_root)
# this call creates a file ``self.dbfn`` if it doesn't exist
sqldb = SQLiteDB(self.dbfn, table=self.db_table)
# max_idx: counter for calc dir numbering
revision = 0
if have_new_db:
max_idx = -1
else:
if mode == 'a':
if sqldb.has_column('idx'):
max_idx = sqldb.execute("select max(idx) from %s" \
%self.db_table).fetchone()[0]
else:
raise StandardError("database '%s': table '%s' has no "
"column 'idx', don't know how to number calcs"
%(self.dbfn, self.db_table))
if sqldb.has_column('revision'):
revision = int(sqldb.get_single("select max(revision) \
from %s" %self.db_table)) + 1
elif mode == 'w':
max_idx = -1
sql_records = []
hostnames = []
for imach, machine in enumerate(self.machines):
hostnames.append(machine.hostname)
calc_dir = pj(self.calc_root, self.calc_dir_prefix + \
'_%s' %machine.hostname)
if os.path.exists(calc_dir):
if backup:
common.backup(calc_dir)
if mode == 'w':
common.system("rm -r %s" %calc_dir, wait=True)
run_txt = "here=$(pwd)\n"
for _idx, params in enumerate(self.params_lst):
params = common.flatten(params)
idx = max_idx + _idx + 1
calc_subdir = pj(calc_dir, str(idx))
extra_dct = \
{'revision': revision,
'study_name': self.study_name,
'idx': idx,
'calc_name' : self.study_name + "_run%i" %idx,
}
extra_params = [SQLEntry(key=key, sqlval=val) for key,val in \
extra_dct.iteritems()]
# templates[:] to copy b/c they may be modified in Calculation
calc = Calculation(machine=machine,
templates=self.templates[:],
params=params + extra_params,
calc_dir=calc_subdir,
)
if mode == 'w' and os.path.exists(calc_subdir):
shutil.rmtree(calc_subdir)
calc.write_input()
run_txt += "cd %i && %s %s && cd $here && sleep %i\n" %(idx,\
machine.subcmd, machine.get_jobfile_basename(), sleep)
if imach == 0:
sql_records.append(calc.get_sql_record())
common.file_write(pj(calc_dir, 'run.sh'), run_txt)
for record in sql_records:
record['hostname'] = SQLEntry(sqlval=','.join(hostnames))
# for incomplete parameters: collect header parts from all records and
# make a set = unique entries
raw_header = [(key, entry.sqltype.upper()) for record in sql_records \
for key, entry in record.iteritems()]
header = list(set(raw_header))
if have_new_db:
sqldb.create_table(header)
else:
for record in sql_records:
for key, entry in record.iteritems():
if not sqldb.has_column(key):
sqldb.add_column(key, entry.sqltype.upper())
for record in sql_records:
cmd = "insert into %s (%s) values (%s)"\
%(self.db_table,
",".join(record.keys()),
",".join(['?']*len(record.keys())))
sqldb.execute(cmd, tuple(entry.sqlval for entry in record.itervalues()))
if excl and revision > 0 and sqldb.has_column('revision'):
old_idx_lst = [str(x) for x, in sqldb.execute("select idx from calc where \
revision < ?", (revision,))]
common.file_write(pj(self.calc_root, 'excl_push'),
'\n'.join(old_idx_lst))
sqldb.finish()
def write_input(self, mode='a', backup=True, sleep=0, excl=True):
"""
Create calculation dir(s) for each parameter set and write input files
based on ``templates``. Write sqlite database storing all relevant
parameters. Write (bash) shell script to start all calculations (run
locally or submitt batch job file, depending on ``machine.subcmd``).
Parameters
----------
mode : str, optional
Fine tune how to write input files (based on ``templates``) to calc
dirs calc_foo/0/, calc_foo/1/, ... . Note that this doesn't change
the base dir calc_foo at all, only the subdirs for each calc.
{'a', 'w'}
| 'a': Append mode (default). If a previous database is found, then
| subsequent calculations are numbered based on the last 'idx'.
| calc_foo/0 # old
| calc_foo/1 # old
| calc_foo/2 # new
| calc_foo/3 # new
| 'w': Write mode. The target dirs are purged and overwritten. Also,
| the database (self.dbfn) is overwritten. Use this to
| iteratively tune your inputs, NOT for working on already
| present results!
| calc_foo/0 # new
| calc_foo/1 # new
backup : bool, optional
Before writing anything, do a backup of self.calc_dir if it already
exists.
sleep : int, optional
For the script to start (submitt) all jobs: time in seconds for the
shell sleep(1) commmand.
excl : bool
If in append mode, a file <calc_root>/excl_push with all indices of
calculations from old revisions is written. Can be used with
``rsync --exclude-from=excl_push`` when pushing appended new
calculations to a cluster.
"""
assert mode in ['a', 'w'], "Unknown mode: '%s'" % mode
if os.path.exists(self.dbfn):
if backup:
common.backup(self.dbfn)
if mode == 'w':
os.remove(self.dbfn)
have_new_db = not os.path.exists(self.dbfn)
common.makedirs(self.calc_root)
# this call creates a file ``self.dbfn`` if it doesn't exist
sqldb = SQLiteDB(self.dbfn, table=self.db_table)
# max_idx: counter for calc dir numbering
revision = 0
if have_new_db:
max_idx = -1
else:
if mode == 'a':
if sqldb.has_column('idx'):
max_idx = sqldb.execute("select max(idx) from %s" \
%self.db_table).fetchone()[0]
else:
raise Exception(
"database '%s': table '%s' has no "
"column 'idx', don't know how to number calcs" %
(self.dbfn, self.db_table))
if sqldb.has_column('revision'):
revision = int(
sqldb.get_single("select max(revision) \
from %s" % self.db_table)) + 1
elif mode == 'w':
max_idx = -1
sql_records = []
hostnames = []
for imach, machine in enumerate(self.machines):
hostnames.append(machine.hostname)
calc_dir = pj(self.calc_root, self.calc_dir_prefix + \
'_%s' %machine.hostname)
if os.path.exists(calc_dir):
if backup:
common.backup(calc_dir)
if mode == 'w':
common.system("rm -r %s" % calc_dir, wait=True)
run_txt = "here=$(pwd)\n"
for _idx, params in enumerate(self.params_lst):
params = common.flatten(params)
idx = max_idx + _idx + 1
calc_subdir = pj(calc_dir, str(idx))
extra_dct = \
{'revision': revision,
'study_name': self.study_name,
'idx': idx,
'calc_name' : self.study_name + "_run%i" %idx,
}
extra_params = [SQLEntry(key=key, sqlval=val) for key,val in \
extra_dct.items()]
# templates[:] to copy b/c they may be modified in Calculation
calc = Calculation(
machine=machine,
templates=self.templates[:],
params=params + extra_params,
calc_dir=calc_subdir,
)
if mode == 'w' and os.path.exists(calc_subdir):
shutil.rmtree(calc_subdir)
calc.write_input()
run_txt += "cd %i && %s %s && cd $here && sleep %i\n" %(idx,\
machine.subcmd, machine.get_jobfile_basename(), sleep)
if imach == 0:
sql_records.append(calc.get_sql_record())
common.file_write(pj(calc_dir, 'run.sh'), run_txt)
for record in sql_records:
record['hostname'] = SQLEntry(sqlval=','.join(hostnames))
# for incomplete parameters: collect header parts from all records and
# make a set = unique entries
raw_header = [(key, entry.sqltype.upper()) for record in sql_records \
for key, entry in record.items()]
header = list(set(raw_header))
if have_new_db:
sqldb.create_table(header)
else:
for record in sql_records:
for key, entry in record.items():
if not sqldb.has_column(key):
sqldb.add_column(key, entry.sqltype.upper())
for record in sql_records:
cmd = "insert into %s (%s) values (%s)"\
%(self.db_table,
",".join(list(record.keys())),
",".join(['?']*len(list(record.keys()))))
sqldb.execute(cmd,
tuple(entry.sqlval for entry in record.values()))
if excl and revision > 0 and sqldb.has_column('revision'):
old_idx_lst = [
str(x) for x, in sqldb.execute(
"select idx from calc where \
revision < ?", (
revision, ))
]
common.file_write(pj(self.calc_root, 'excl_push'),
'\n'.join(old_idx_lst))
sqldb.finish()
cell_c_min = cell_c.min()
etot = np.array([(a-cell_a_mean)**2.0 + (c-cell_c_mean)**2.0 for a,c in axes_flat])
phdos = []
Vmax = V.max()
# phonon dos (just a gaussian) shifted to lower (higher) freqs for higher
# (lower) volume
for ii in range(axes_flat.shape[0]):
a,c = axes_flat[ii,:]
fc = 550 - 50 * V[ii] / Vmax
phdos.append(np.array([freq,gauss(freq-fc,100)*0.01]).T)
gibbs = Gibbs(T=T, P=P, etot=etot, phdos=phdos, axes_flat=axes_flat,
volfunc_ax=volfunc_ax, case=case, dosarea=None)
gibbs.set_fitfunc('C', lambda x,y: num.Spline(x,y,s=None,k=5, eps=1e-5))
g = gibbs.calc_G(calc_all=True)
common.makedirs('../files/gibbs/2d')
io.write_h5('../files/gibbs/2d/%s.h5' %gethostname(), filt_dct(g), mode='w')
# 1d case
case = '1d'
V = np.linspace(10,20,nax)
axes_flat = V**(1/3.) # cubic
volfunc_ax = lambda x: x[0]**3.0
etot = (V-V.mean())**2
fcenter = 450 + 100*(axes_flat - axes_flat.min())
# fake phonon dos data (Gaussian), shift to lower freq for higher volume
phdos = [np.array([freq,gauss(freq-fc, 100)]).T for fc in
fcenter[::-1]]
gibbs = Gibbs(T=T, P=P, etot=etot, phdos=phdos, axes_flat=axes_flat,