def iterate(self, method="hessian", supercell=None, q=None, nrattle=0):
"""Returns a list of possible configurations, one for each eigenmode in the
system, that has `supercell` is compatible with the specified `q`
vector.
Args:
method (str): desired method for computing the eigenvalues and
eigenvectors. Possible choices are :meth:`hessian` or
:meth:`dynmat`.
supercell (numpy.ndarray): supercell matrix to use in generating the
configs.
q (numpy.ndarray): q-vector that the resulting supercell should be
compatible with.
nrattle (int): number of additional, empty configs to include via
:meth:`quippy.Atoms.rattle`.
"""
if method == "hessian":
dmd = self.hessian()
elif method == "vasp_hessian":
dmd = self.vasp_hessian()
else:
dmd = self.dynmat(supercell, q)
hname = "hessian1"
seed = quippy.Atoms()
seed.copy_from(dmd["template"])
result = quippy.AtomsList()
result.append(dmd["template"])
#Delete the force, energy and virial information from the copy, so that
#we don't duplicate it all over.
del seed.params["dft_energy"]
del seed.params["dft_virial"]
del seed.properties["dft_force"]
for l, v in zip(dmd["eigvals"], dmd["eigvecs"]):
atc = seed.copy()
#Add this eigenvector to its own configuration.
atc.add_property(hname, 0.0, n_cols=3)
H = np.reshape(v.real, (atc.n, 3)).T
setattr(atc, hname, H)
#Same thing for the eigenvalue. Then save it to the group folder
#structure.
atc.params.set_value(hname, l)
atc.params.set_value("n_hessian", 1)
#atc.add_property("force", 0.0, n_cols=3)
result.append(atc)
for i in range(nrattle):
atRand = seed.copy()
quippy.randomise(atRand.pos, 0.2)
result.append(atRand)
# atz = seed.copy()
# atz.add_property("dft_force", 0.0, n_cols=3)
#atz.params.set_value("dft_energy",
return result
def main(suffix, fxyz, rc, species, nmax, lmax, awidth, nframes, cutoff_dexp,
cutoff_scale):
suffix = str(suffix)
fxyz = str(fxyz)
cutoff = float(rc)
species = sorted([int(species) for species in species.split(',')])
nmax = int(nmax)
lmax = int(lmax)
awidth = float(awidth)
nframes = int(nframes)
if nframes == 0: nframes = None
nspecies = len(species)
nn = nmax**2 * (lmax + 1)
nab = nspecies * (nspecies + 1) / 2
cutoff_dexp = int(cutoff_dexp)
cutoff_scale = float(cutoff_scale)
frames = qp.AtomsList(fxyz, stop=nframes)
nframes = len(frames)
soapstr = Template('average=F normalise=T soap cutoff_dexp=$cutoff_dexp \
cutoff_scale=$cutoff_scale central_reference_all_species=F \
central_weight=1.0 covariance_sigma0=0.0 atom_sigma=$awidth \
cutoff=$rc cutoff_transition_width=0.5 n_max=$nmax l_max=$lmax \
n_species=$nspecies species_Z=$species n_Z=$ncentres Z=$centres'
)
soapstr = soapstr.safe_substitute(rc=rc, nmax=nmax, lmax=lmax, awidth=awidth, \
cutoff_dexp=cutoff_dexp, cutoff_scale=cutoff_scale)
soapstr = Template(soapstr)
gsoaps = get_soaps(soapstr, rc, species, nspecies, nmax, lmax, nn, nab)
soaps_list = gsoaps(frames)
p = {}
for i in xrange(nframes):
p[i] = soaps_list[i]
f = open(suffix + '.pckl', 'wb')
pickle.dump(p, f)
f.close()
def main(filename, nmax, lmax, coff, cotw, gs, centerweight, prefix=""):
start_time = datetime.now()
filename = filename[0]
# sets a few defaults
if prefix == "": prefix = filename
if prefix.endswith('.xyz'): prefix = prefix[:-4]
prefix = prefix + "-n" + str(nmax) + "-l" + str(lmax) + "-c" + str(
coff) + "-g" + str(gs)
print >> sys.stderr, "using output prefix =", prefix
# Reads input file using quippy
print >> sys.stderr, "Reading input file", filename
(first, last) = (0, 0)
# reads the whole thing
if first == 0: first = None
if last == 0: last = None
al = quippy.AtomsList(filename, start=first, stop=last)
print >> sys.stderr, len(al.n), " Configurations Read"
# determines "kit" (i.e. max n and kind of atoms present)
spkit = {}
for at in al:
atspecies = {}
for z in at.z:
if z in atspecies: atspecies[z] += 1
else: atspecies[z] = 1
for (z, nz) in atspecies.iteritems():
if z in spkit:
if nz > spkit[z]: spkit[z] = nz
else:
spkit[z] = nz
# species string
zsp = spkit.keys()
zsp.sort()
lspecies = 'n_species=' + str(len(zsp)) + ' species_Z={ '
for z in zsp:
lspecies = lspecies + str(z) + ' '
lspecies = lspecies + '}'
print lspecies
fout = open(prefix + ".soap", "w")
for at in al:
fout.write("%d\n" % (len(at.z)))
at.set_cutoff(coff)
at.calc_connect()
nel = np.bincount(at.z)
# ok, it appears you *have* to do this atom by atom. let's do that, but then re-sort in the same way as in the input
soaps = {}
sz = {}
for z in at.z:
if z in sz: sz[z] += 1
else: sz[z] = 1
for (z, nz) in sz.iteritems():
soapstr = ("soap central_reference_all_species=F central_weight=" +
str(centerweight) +
" covariance_sigma0=0.0 atom_sigma=" + str(gs) +
" cutoff=" + str(coff) + " cutoff_transition_width=" +
str(cotw) + " n_max=" + str(nmax) + " l_max=" +
str(lmax) + ' ' + lspecies + ' Z=' + str(z))
desc = quippy.descriptors.Descriptor(soapstr)
print soapstr
soaps[z] = desc.calc(at)["descriptor"]
for z in at.z:
fout.write("%3s " % (atomicno_to_sym(z)))
np.savetxt(fout, [soaps[z][len(soaps[z]) - sz[z]]])
sz[z] -= 1
print sz
fout.close()
# Wait for instructions from the master process.
pool.wait()
sys.exit(0)
print_logo()
print "Start dirty parallelisation for cluster " \
"with a pool of {} mpi processes: {}".format(pool.size,ctime())
print "Computing the global {} kernel of {} " \
"from index {} to {} (None -> default, i.e. begining/end)".format(global_kernel_type, filename,first,last)
# run commands in parallel
st = time()
frames = qp.AtomsList(filename,start=first,stop=last)
Nframe = len(frames)
if args.chunklen:
aa = args.chunklen.split(',')
if len(aa) > 1:
xchunklen, ychunklen = int(aa[0]), int(aa[1])
elif len(aa) == 1:
xchunklen, ychunklen = int(aa[0]), int(aa[0])
else:
raise Exception('chunk lenght format not recognised')
else:
print 'Default chunk lenght'
xchunklen, ychunklen = Nframe // 5, Nframe // 5
xNchunk = Nframe // xchunklen
# format of the output name
if prefix == "": prefix = filename
if prefix.endswith('.xyz'): prefix = prefix[:-4]
prefix += "-n" + str(nmax) + "-l" + str(lmax) + "-c" + str(cutoff) + \
"-g" + str(gaussian_width) + "-cw" + str(centerweight) + \
"-cotw" + str(cutoff_transition_width)
fn_env_kernels = prefix + '-env_kernels.pck'
print "using output prefix =", prefix
st = time.time()
################################################################################
# Reads the file and create a list of quippy frames object
atoms1 = qp.AtomsList(filename, start=xlim[0], stop=xlim[1])
print 'Reading {} input atomic structure from {} with index {}: done {}'.format(
len(atoms1), filename, xlim, s2hms(time.time() - st))
if xlim != ylim:
atoms2 = qp.AtomsList(filename, start=ylim[0], stop=ylim[1])
print 'Reading {} input atomic structure from {} with index {}: done {}'.format(
len(atoms2), filename, ylim, s2hms(time.time() - st))
else:
atoms2 = None
print 'no atoms 2, Computing upper triangular sub matrix'
soap_params = {
'centerweight': centerweight,
'gaussian_width': gaussian_width,
'cutoff': cutoff,
'cutoff_transition_width': cutoff_transition_width,
def main(filename, nmax, lmax, coff, cotw, gs, centerweight, gspecies, redfiles, prefix=""):
start_time = datetime.now()
filename = filename[0]
# sets a few defaults
if prefix=="": prefix=filename
if prefix.endswith('.xyz'): prefix=prefix[:-4]
prefix=prefix+"-n"+str(nmax)+"-l"+str(lmax)+"-c"+str(coff)+"-g"+str(gs)
print >> sys.stderr, "using output prefix =", prefix
# Reads input file using quippy
print >> sys.stderr, "Reading input file", filename
(first,last)=(0,0); # reads the whole thing
if first==0: first=None;
if last==0: last=None
al = quippy.AtomsList(filename, start=first, stop=last);
print >> sys.stderr, len(al.n) , " Configurations Read"
# determines "kit" (i.e. max n and kind of atoms present)
spkit = {}
for at in al:
atspecies = {}
for z in at.z:
if z in atspecies: atspecies[z]+=1
else: atspecies[z] = 1
for (z, nz) in atspecies.iteritems():
if z in spkit:
if nz>spkit[z]: spkit[z] = nz
else:
spkit[z] = nz
# species string
if gspecies is None:
zsp=spkit.keys()
zsp.sort()
print "No species provided"
else: zsp=map(int,gspecies.split(","))
# If dimensionality reduction is called, stores the variables
cur = False
if redfiles is not None:
cur = True
# Does not matter in which sequence the files are provided
try:
idxsel = np.loadtxt(redfiles[0], dtype=int)
chol = np.loadtxt(redfiles[1])
except ValueError:
idxsel = np.loadtxt(redfiles[1], dtype=int)
chol = np.loadtxt(redfiles[0])
# Checks if all dimensions are the same
assert idxsel.shape[0] == chol.shape[0] == chol.shape[1], "Wrong dimensions, cannot perform dimensionality reduction"
lspecies = 'n_species='+str(len(zsp))+' species_Z={ '
for z in zsp: lspecies = lspecies + str(z) + ' '
lspecies = lspecies + '}'
print "Using the following species:", ",".join([str(z) for z in zsp])
if not cur:
fout=open(prefix+".soap","w")
else:
fout=open(prefix+"_red.soap","w") # To avoid giving the same name to the reduced soap vector
counter = 0
for at in al:
fout.write("%d\n" % (len(at.z)))
at.set_cutoff(coff);
at.calc_connect();
nel = np.bincount(at.z)
# ok, it appears you *have* to do this atom by atom. let's do that, but then re-sort in the same way as in the input
soaps = {}
sz = {}
for z in at.z:
if z in sz: sz[z]+=1
else: sz[z]=1
for (z, nz) in sz.iteritems():
soapstr=("soap central_reference_all_species=F central_weight="+str(centerweight)+
" covariance_sigma0=0.0 atom_sigma="+str(gs)+" cutoff="+str(coff)+" cutoff_transition_width="+str(cotw)+
" n_max="+str(nmax)+" l_max="+str(lmax)+' '+lspecies+' Z='+str(z))
desc = quippy.descriptors.Descriptor(soapstr )
soaps[z] = desc.calc(at)["descriptor"]
for z in at.z:
fout.write("%3s " % (atomicno_to_sym(z)))
if cur:
single_soap = soaps[z][len(soaps[z])-sz[z]]
red_soap = np.dot(single_soap[idxsel], chol)
np.savetxt(fout, [red_soap])
else: np.savetxt(fout, [ soaps[z][len(soaps[z])-sz[z]] ])
sz[z] -=1
counter +=1
sys.stderr.write("SOAP vectors calculated: %d / %d \r" %(counter, len(al.n)) )
sys.stderr.write("\n")
fout.close()
default=1.0 * units.fs)
parser.add_argument('-dt_traj',
'--traj_interval',
help='Trajectory write frequency',
type=int,
default=10)
# ***** End input parameters *****
params = parser.parse_args()
calc = quippy.Potential('IP TS', param_filename="ts_params.xml")
tipatoms = params.tipatoms
folder, config_filename = os.path.split(params.config_filename)
os.chdir(folder)
at = quippy.AtomsList(config_filename, start=-1)[0]
# automatically set spring constant to default Gc value if not manually set
if params.k_spring == -1:
if 'c' in folder:
params.k_spring = 0.35
elif 'a' in folder:
params.k_spring = 0.80
print("Inferred k_spring %f" % params.k_spring)
else:
print("Please specify the spring constant.")
sys.exit(1)
print("Running MD with following parameters:")
print(params)
def main(fnamexyz,
cutoffdist,
fnamesim,
zenv,
nenv,
dcut,
selectgroupxyz2print=[],
dcutisgood=False,
isploting=False):
pyplot = isploting
plotdendo = True
# Generate the file name for the new dist matrices that are without dummy atoms
nsim = len(nenv)
fname = []
fnamecorr = []
fnamemathematica = []
cnt = 0
for fn in fnamesim:
target = '_' + str(nenv[cnt])
fname.append(fn[:fn.find(target)] + '.sim')
fnamecorr.append(fn[:fn.find(target)] + '.corrmt')
fnamemathematica.append(fn[:fn.find(target)] +
'.mathematica-cluster.dat')
cnt += 1
print(fn[:fn.find(target)])
print >> sys.stderr, "Reading input file", fnamexyz
mols = quippy.AtomsList(fnamexyz)
print >> sys.stderr, len(mols.n), " Configurations Read"
distmatrix = []
clist = []
nbclst = []
clusterlist = []
nspiecies = []
Z = []
dendo = []
for it in range(nsim):
print '############'
print "Read distance matrix " + fnamesim[it] + " with dummy atoms"
olddistmatrix = np.loadtxt(fnamesim[it])
fsim = open(fnamesim[it], 'r')
head = fsim.readline()
head = head.strip('#')
head = head.strip('\n')
# print '##########' , head
fsim.close()
# Removes dummy atoms rows/columns from distance matrix
newdistmatrix1, dummylist = env_corr.rmdummyfromsim(
fnamexyz, olddistmatrix, zenv[it], nenv[it])
print len(newdistmatrix1), ' Real ', zenv[it], ' atom '
print 'Removes : ', len(dummylist), ' rows/colomns from distmatrix'
print newdistmatrix1.shape
# Write the distmatrix without dummy atoms
np.savetxt(fname[it], newdistmatrix1, header=head)
# get list of the cluster groups idx (same order as in dist mat)
clist1, Z1 = env_corr.clusterdistmat(fname[it],
newdistmatrix1,
0.,
mode='average',
plot=plotdendo)
dendo1 = sc.dendrogram(Z1)
# Write mathematica cluster structure
cluster.mathematica_cluster(Z1, newdistmatrix1, fnamemathematica[it])
# get nb of cluster groups
nbclst1 = len(np.unique(clist1))
# print a==nbclst1
distmatrix.append(newdistmatrix1)
clist.append(clist1)
nbclst.append(nbclst1)
Z.append(Z1)
dendo.append(dendo1)
# Link the cluster groups with atom's frame and respective position in it
if dcutisgood:
clusterlist1, nspiecies1 = env_corr.linkgroup2atmidx(
mols, clist1, zenv[it], nenv[it])
clusterlist.append(clusterlist1)
# count nb of species atom in each frame
nspiecies.append(nspiecies1)
# Compute the correlation between the species of 2 cluster
# look for the groups of the atoms (from cluster2) in the environment of an
# atom from a group from cluster1
if dcutisgood:
corrmtx = []
for it in range(1, nsim):
if len(selectgroupxyz2print) == 0:
selectgroupxyz2print = []
for i in range(1, nsim + 1):
selectgroupxyz2print.append([])
corrmtx1, corr1, idx1, idx2 = env_corr.getcorrofcluster(
mols, cutoffdist, fnamexyz, clusterlist[0], clusterlist[it],
nbclst[0], nbclst[it], zenv[0], zenv[it], nspiecies[0],
nspiecies[it], selectgroupxyz2print[it - 1])
# idx1 = dendo[0]['leaves']
# idx2 = dendo[it]['leaves']
print len(idx1), len(idx2)
print corrmtx1.shape
# corrmtx1 = corrmtx1[idx1,:]
# corrmtx1 = corrmtx1[:,idx2]
corrmtx.append(corrmtx1)
head1 = 'From ' + fnamexyz + ' Correlation matrix between environment of atom ' + str(
zenv[0]) + ' and ' + str(zenv[it])
print '############'
print head1
head2 = 'From ' + fnamexyz + ' Dendogram ordering of atom ' + str(
zenv[0])
head3 = 'From ' + fnamexyz + ' Dendogram ordering of atom ' + str(
zenv[it])
np.savetxt(fnamecorr[it], corrmtx1, header=head1)
np.savetxt(fnamecorr[it] + str(zenv[0]), idx1, header=head2)
np.savetxt(fnamecorr[it] + str(zenv[it]), idx2, header=head3)
# print corrmtx1
if pyplot:
plt.show()
debug = True
save_traj = True
n_steps_heat = 20000
n_steps_equil = 10000
n_steps_liquid = 5000
T_melt = 5000.0
T_equil = 2000.0
liquid_traj_filename = "model-" + model.name + "-test-liquid-liquid_traj.xyz"
if __builtin__.do_io:
print liquid_traj_filename
try:
liquid_traj = quippy.AtomsList(liquid_traj_filename)
if __builtin__.do_io:
print "read %s from file" % liquid_traj_filename
except:
if __builtin__.do_io:
print "generating %s" % liquid_traj_filename
if 'LIQUID_NO_RUN' in os.environ:
sys.exit(1)
# specify that we will use model.calculator to compute forces, energies and stresses
bulk.set_calculator(model.calculator)
print "doing bulk"
print bulk.get_potential_energy()
print "done"
try:
model.calculator.print_()
def main():
parser = argparse.ArgumentParser(
description="""Computes the similarity matrix between a set of atomic structures based on SOAP descriptors and an optimal assignment of local environments.""")
parser.add_argument("filename", nargs=1, help="Name of the formatted xyz input file")
parser.add_argument("--kit", type=str, default="None", help="Dictionary-style kit specification (e.g. --kit '{4:1,6:10}' or 'auto' or 'None' (default)")
parser.add_argument("--continue_from", type=str, default=None, help="Load kernel and compute missing (nan) entries")
args = parser.parse_args()
with open("alchemy.pickle","rb") as alchem_file:
alchem_rules = pickle.load(alchem_file)
#from collections import defaultdict
#alchem_rules = defaultdict(lambda: 1) # TODO: WARNING. Using 1 as alchem rule
#print "ALCHEM RULES ARE 1"
xyz_file = args.filename[0]
kernel_file = os.path.splitext(xyz_file)[0] + ".k.pkz"
# Load xyz file with molecules and convert to quippy format
all_mols = quippy.AtomsList(xyz_file)
# Determine kit to use
kitstr = args.kit
if kitstr == "auto":
kit = getkit(all_mols)
print "Using kit %s" % kit
elif kitstr != "None" and kitstr != "none":
kit = ast.literal_eval(str(args.kit))
print "Using kit %s" % kit
else:
kit = None
# Compute descriptor for each molecule
print "Computing descriptors"
descriptors = []
for mol in all_mols:
descriptors.append(PickableMol(SoapMol(alchem_rules, mol, kit = kit)))
# Create kernel matrix
N = len(all_mols)
if args.continue_from is None:
kernel_matrix = np.ones((N, N)) * np.nan
else:
with gzip.open(args.continue_from, "rb") as f:
kernel_matrix = pickle.load(f)
assert kernel_matrix.shape == (N,N), "Can't continue, loaded kernel has wrong dimension"
assert not np.any(np.isnan(np.diag(kernel_matrix))), "Can't continue, diagonal has missing elements"
assert not np.any(np.diag(kernel_matrix) == 1.0), "Can't continue if diagonal already normalised"
# Setup multiprocessing
job_queue = multiprocessing.Queue()
res_queue = multiprocessing.Queue()
num_processes = 4
chunk_size = 100
usekit = (kit is not None)
if usekit:
alchem = get_alchem_mat(alchem_rules, kit)
else:
alchem = alchem_rules
workers = [
multiprocessing.Process(
target=worker, args=(job_queue, res_queue, alchem, descriptors, usekit)) \
for _ in range(num_processes)
]
# Start all workers
for w in workers:
w.start()
# First compute unnormalised diagonal (used for normalisation)
if args.continue_from is None:
print "Computing diagonal"
for (row, col) in zip(range(N), range(N)):
job_queue.put([(row, col),])
# Retrieve results
num_jobs_finished = 0
while num_jobs_finished < N:
row, col, res = res_queue.get()
kernel_matrix[row,col] = res
num_jobs_finished += 1
else:
print "Diagonal already computed, skipping"
# Compute off-diagonal elements (with normalisation)
print "Computing off-diagonal elements"
# Submit jobs to queue
if args.continue_from is None:
all_jobs = list(itertools.combinations(range(N),2))
else:
unfinished = np.transpose(np.nonzero(np.isnan(kernel_matrix)))
all_jobs = [(row, col) for row, col in unfinished if row<col]
for chunk in grouper(chunk_size, all_jobs, (None, None)):
job_queue.put(chunk)
# Retrieve results
num_jobs_finished = 0
total_jobs = len(all_jobs)
while num_jobs_finished < total_jobs:
row, col, res = res_queue.get()
inorm = 1.0/np.sqrt(kernel_matrix[row,row]*kernel_matrix[col,col])
kernel_matrix[row,col] = res*inorm
kernel_matrix[col,row] = res*inorm
num_jobs_finished += 1
if (num_jobs_finished % 10000) == 0:
print "%s job %d/%d complete (%d%%)" %\
(time.strftime("%Y-%m-%d %H:%M"),
num_jobs_finished,
total_jobs,
100.0*float(num_jobs_finished)/float(total_jobs))
if (num_jobs_finished % 1000000) == 0:
# Save kernel matrix to file
with gzip.open(kernel_file, "wb") as f:
pickle.dump(kernel_matrix, f)
print "Done computing off-diagonal, saving kernel"
# Diagonal is normalised to unity
kernel_matrix[np.diag_indices(N)] = 1.0
# Save kernel matrix to file
with gzip.open(kernel_file, "wb") as f:
pickle.dump(kernel_matrix, f)
print "Joining workers"
for w in workers:
job_queue.put([(None, None),])
for w in workers:
w.join()
