def run(self, out):
Nit = len(self.chunks)
pbar = tqdm_cs(total=Nit,
desc='SOAP vectors',
disable=self.disable_pbar)
results = out
ids_list = self.ids_list
if self.nprocess > 1:
pool = mp.Pool(self.nprocess,
initializer=self.worker_init,
maxtasksperchild=10)
for idx, res in pool.imap_unordered(self.func_wrap, self.chunks):
results[ids_list[idx], :] = res
pbar.update()
pool.close()
pool.join()
elif self.nprocess == 1:
for chunk in self.chunks:
idx, res = self.func_wrap(chunk)
results[ids_list[idx], :] = res
pbar.update()
else:
print 'Nproces: ', self.nprocess
raise NotImplementedError('need at least 1 process')
pbar.close()
def run(self):
Nit = len(self.chunks)
pbar = tqdm_cs(total=Nit,desc='SOAP vectors',disable=self.dispbar)
results = {}
if self.nprocess > 1:
pool = mp.Pool(self.nprocess, initializer=self.worker_init,
maxtasksperchild=10)
for idx, res in pool.imap_unordered(self.func_wrap, self.chunks):
results[idx] = res
pbar.update()
pool.close()
pool.join()
elif self.nprocess == 1:
for chunk in self.chunks:
idx,res = self.func_wrap(chunk)
results[idx] = res
pbar.update()
else:
print('Nproces: ',self.nprocess)
raise NotImplementedError('need at least 1 process')
pbar.close()
Frames = [results[it] for it in range(Nit)]
return Frames
def run(self):
Nit = len(self.chunks)
pbar = tqdm_cs(total=Nit, desc='LearningCurve', disable=self.dispbar)
scores_test = []
scores_train = []
scores_test_err = []
scores_train_err = []
Ntrains = []
if self.nprocess > 1:
pool = Pool(self.nprocess,
initializer=self.worker_init,
maxtasksperchild=1)
for Ntrain, res_test, res_train in pool.imap_unordered(
self.func_wrap, self.chunks):
scores_test.append(np.mean(res_test, axis=0))
scores_train.append(np.mean(res_train, axis=0))
scores_test_err.append(np.std(res_test, axis=0))
scores_train_err.append(np.std(res_train, axis=0))
Ntrains.append(Ntrain)
pbar.update()
pool.close()
pool.join()
elif self.nprocess == 1:
for chunk in self.chunks:
Ntrain, res_test, res_train = self.func_wrap(chunk)
scores_test.append(np.mean(res_test, axis=0))
scores_train.append(np.mean(res_train, axis=0))
scores_test_err.append(np.std(res_test, axis=0))
scores_train_err.append(np.std(res_train, axis=0))
Ntrains.append(Ntrain)
pbar.update()
else:
print 'Nproces: ', self.nprocess
raise NotImplementedError('need at least 1 process')
pbar.close()
scores_test = np.asarray(scores_test)
scores_train = np.asarray(scores_train)
scores_test_err = np.asarray(scores_test_err)
scores_train_err = np.asarray(scores_train_err)
return Ntrains, scores_test, scores_test_err, scores_train, scores_train_err
def run(self):
Nit = len(self.chunks)
pbar = tqdm_cs(total=Nit, desc='GridSearch', disable=self.dispbar)
scores = []
gs_params = []
if self.nprocess > 1:
pool = Pool(self.nprocess,
initializer=self.worker_init,
maxtasksperchild=1)
for param, res in pool.imap_unordered(self.func_wrap, self.chunks):
scores.append(res)
gs_params.append(param)
pbar.update()
pool.close()
pool.join()
elif self.nprocess == 1:
for chunk in self.chunks:
param, res = self.func_wrap(chunk)
scores.append(res)
gs_params.append(param)
pbar.update()
else:
print 'Nproces: ', self.nprocess
raise NotImplementedError('need at least 1 process')
pbar.close()
scores = np.asarray(scores)
ii = np.argmin(scores[:, 0])
bestParams = gs_params[ii]
bestParams.pop('memory_eff')
return bestParams, scores, gs_params
def __init__(self, Niter, name, dispbar):
# super(dummy_queue,self).__init__()
self.tbar = tqdm_cs(total=int(Niter), desc=name, disable=dispbar)
def simulateur_NVT_efficace(positions,
velocities,
masse,
temperature,
r_m,
epsilon,
k_spring,
Nstep,
dt,
enregistrement_stride=10):
mass = masse
Nparticule, _ = positions.shape
accelerations = np.zeros(positions.shape)
pos = []
vel = []
masses = np.ones((Nparticule)) * mass
thermostat_frequency = np.sqrt(2) * np.sqrt(2 * k_spring)
Nrecord = Nstep // enregistrement_stride
diagnostic = dict(E_variation=np.zeros((Nrecord, )),
Temperature=np.zeros((Nrecord, )),
E_system=np.zeros((Nrecord, )),
E_potentiel=np.zeros((Nrecord, )),
E_cinetique=np.zeros((Nrecord, )),
time=np.zeros((Nrecord, )))
dt_half = 0.5 * dt
#th_en = 0.5*mass*np.power(velocities,2).sum()
sys_en = 0
econs = 0.0
# Calculates the initial potential energy
forces, ljpot, sppot = get_forces(positions, r_m, epsilon, k_spring)
Ekin_tot = 0.5 * mass * np.power(velocities, 2).sum()
sys_en += ljpot + sppot + Ekin_tot
econs += sys_en
ii = 0
for it in tqdm_cs(range(Nstep)):
#Apply thermostat
econs += 0.5 * mass * np.power(velocities, 2).sum()
velocities = andersen_thermostat(velocities, temperature,
thermostat_frequency, dt)
#CoM = np.average(velocities,weights=masses,axis=0).reshape((1,3))
#velocities = velocities - CoM
econs -= 0.5 * mass * np.power(velocities, 2).sum()
# half update of velocities
velocities = velocities + dt_half * accelerations
# update of positions
positions = positions + dt * velocities
# update forces from new posittions
forces, ljpot, sppot = get_forces(positions, r_m, epsilon, k_spring)
# update acceleration
accelerations = forces / mass
# half update of velocities
velocities = velocities + dt_half * accelerations
Ekin_tot = 0.5 * mass * np.power(velocities, 2).sum()
econs += ljpot + sppot + Ekin_tot - sys_en
sys_en = ljpot + sppot + Ekin_tot
if it % enregistrement_stride == 0:
diagnostic['E_variation'][ii] = econs
diagnostic['E_system'][ii] = sys_en
diagnostic['E_cinetique'][ii] = Ekin_tot
diagnostic['time'][ii] = it * dt
diagnostic['Temperature'][ii] = Ekin_tot / 1.5 / len(positions)
diagnostic['E_potentiel'][ii] = ljpot + sppot
ii += 1
vel.append(velocities)
pos.append(positions)
return pos, vel, diagnostic
def listener(queue, Niter, dispbar):
print 'listener ', dispbar
tbar = tqdm_cs(total=int(Niter), desc='Env kernels', disable=dispbar)
for ii in iter(queue.get, None):
tbar.update(ii)
tbar.close()