def compute_toverlaps_in_parallel( step ):
"""
Function used for the making the computation of time-overlaps parallel via python multiprocessing
"""
st_sd = step3.mapping.ovlp_mat_arb( sd_states_reindexed_sorted[step], sd_states_reindexed_sorted[step+1], St_ks[0][step], use_minimal=False )
st_sd = data_conv.MATRIX2nparray(st_sd)
return st_sd
def run_nbra_namd_wrapper(hvib, istate, outfile_name, params):
"""
A small wrapper function to run the nbra namd dynamics.
hvib ( list of list of matrices ): The vibronic hamiltonian for all timesteps
istate ( int ): Index of the initial state. This index is from 0
outfile ( string ): The name of the output file
params ( dict ): A dictionary of important dynamical parameters
returns: A list of energy values that is the electronic energy vs time. Returned energy is in eV
"""
print("decoherence_method = ", params["decoherence_method"])
params["outfile"] = outfile_name
params["nstates"] = hvib[0].num_of_rows
params["istate"] = istate
res_nbra_namd = step4.run([hvib], params)
nbra_namd = data_conv.MATRIX2nparray(res_nbra_namd)
# For SH energies
energy_nbra_namd = (nbra_namd[:, 3 * params["nstates"] + 2] -
nbra_namd[:, 1]) * units.au2ev
return energy_nbra_namd
def run_bllz_wrapper( hvib, istate, outfile_name, params ):
"""
A small wrapper function to run the bllz dynamics.
hvib ( list of list of matrices ): The vibronic hamiltonian for all timesteps
istate ( int ): Index of the initial state. This index is from 0
outfile ( string ): The name of the output file
params ( dict ): A dictionary of important dynamical parameters
returns: A list of energy values that is the electronic energy vs time. Returned energy is in eV
"""
params["outfile"] = outfile_name
params["nstates"] = hvib[0].num_of_rows
params["istate"] = istate
res_bllz, P = lz.run( [hvib], params )
bllz_namd = data_conv.MATRIX2nparray( res_bllz )
# Recall that for bllz, we use the schrodiger not the surface hopping energy
energy_bllz = ( bllz_namd[:,3*params["nstates"]+1] - bllz_namd[:,1] )*units.au2ev
# for bllz surface hopping energy
#energy_bllz = ( bllz_namd[:,3*params["nstates"]+2] - bllz_namd[:,1] )*units.au2ev
return energy_bllz
def run_cp2k_xtb_step2(params):
"""
This function runs the step2 for computing the MO overlaps for xTB calculations and saves them as sparse
matrices using scipy.sparse library. In order to load the saved sparse matrices you need to use
scipy.sparse.load_npz command.
Args:
params (dictionary): A dictionary containing the following parameters:
res_dir (string): The directory for saving the MO overlaps.
all_logfiles (string): The directory to save all log files.
all_pdosfiles (string): The directory to save all pdos files.
istep (integer): The initial step.
fstep (integer): The final step.
init_state (integer): The initial state.
final_state (integer): The final state.
is_spherical (bool): Flag for spherical coordinates.
remove_molden (bool): Flag for removing the molden files after computing the MO overlaps.
nprocs (integer): Number of processors.
cp2k_ot_input_template (string): The CP2K OT input template file name.
cp2k_diag_input_template (string): The CP2K diagonalization input template file name.
trajectory_xyz_filename (string): The trajectory xyz file name.
step (integer): The time step.
cp2k_exe (string): The full path to CP2K executable.
"""
# Making the required directories for gatherig all the information needed
# including the overlap data and pdos files. The logfiles do not contain
# specific data but we finally move them to all_logfiles
if not os.path.exists(params['res_dir']):
os.mkdir(params['res_dir'])
if not os.path.exists(params['all_logfiles']):
os.mkdir(params['all_logfiles'])
if not os.path.exists(params['all_pdosfiles']):
os.mkdir(params['all_pdosfiles'])
# setting up the initial step and final step
istep = params['istep']
fstep = params['fstep']
# spherical or cartesian GTOs
is_spherical = params['is_spherical']
# number of processors
nprocs = params['nprocs']
# the counter for a job steps, this counter is needed for not reading
# the data of a molden file twice
counter = 0
print('-----------------------Start-----------------------')
for step in range(istep, fstep):
# a timer for all the procedure
t1_all = time.time()
print('-----------------------Step %d-----------------------' % step)
params['step'] = step
# timer for CP2K
t1 = time.time()
CP2K_methods.run_cp2k_xtb(params)
print('Done with step', step, 'Elapsed time:', time.time() - t1)
# now if the counter is equal to zero
# just compute the MO overlap of that step.
if counter == 0:
t1 = time.time()
print('Creating shell...')
shell_1, l_vals = molden_methods.molden_file_to_libint_shell('Diag_%d-libra-1_0.molden'%step,\
is_spherical)
print('Done with creating shell. Elapsed time:', time.time() - t1)
t1 = time.time()
print('Reading energies and eigenvectors....')
eig_vect_1, energies_1 = molden_methods.eigenvectors_molden('Diag_%d-libra-1_0.molden'%step,\
nbasis(shell_1),l_vals)
print('Done with reading energies and eigenvectors. Elapsed time:',
time.time() - t1)
if params['remove_molden']:
os.system('rm Diag_%d-libra-1_0.molden' % step)
print('Computing atomic orbital overlap matrix...')
t1 = time.time()
AO_S = compute_overlaps(shell_1, shell_1, nprocs)
print('Done with computing atomic orbital overlaps. Elapsed time:',
time.time() - t1)
t1 = time.time()
print('Turning the MATRIX to numpy array...')
AO_S = data_conv.MATRIX2nparray(AO_S)
print('Done with transforming MATRIX 2 numpy array. Elapsed time:',
time.time() - t1)
istate = params['init_state']
fstate = params['final_state']
## Now, we need to resort the eigenvectors based on the new indices
print('Resorting eigenvectors elements...')
t1 = time.time()
new_indices = CP2K_methods.resort_molog_eigenvectors(l_vals)
eigenvectors_1 = []
for j in range(len(eig_vect_1)):
# the new and sorted eigenvector
eigenvector_1 = eig_vect_1[j]
eigenvector_1 = eigenvector_1[new_indices]
# append it to the eigenvectors list
eigenvectors_1.append(eigenvector_1)
eigenvectors_1 = np.array(eigenvectors_1)
print('Done with resorting eigenvectors elements. Elapsed time:',
time.time() - t1)
##
t1 = time.time()
print('Computing and saving molecular orbital overlaps...')
# Note that we choose the data from istate to fstate
# the values for istate and fstate start from 1
S = np.linalg.multi_dot([eigenvectors_1, AO_S,
eigenvectors_1.T])[istate - 1:fstate - 1,
istate - 1:fstate - 1]
# creating zero matrix
mat_block_size = len(S)
zero_mat = np.zeros((mat_block_size, mat_block_size))
S_step = data_conv.form_block_matrix(S, zero_mat, zero_mat, S)
# Since a lot of the data are zeros we save them as sparse matrices
S_step_sparse = scipy.sparse.csc_matrix(S_step)
E_step = data_conv.form_block_matrix(np.diag(energies_1)[istate-1:fstate-1,istate-1:fstate-1],\
zero_mat,zero_mat,\
np.diag(energies_1)[istate-1:fstate-1,istate-1:fstate-1])
E_step_sparse = scipy.sparse.csc_matrix(E_step)
scipy.sparse.save_npz(params['res_dir'] + '/S_ks_%d.npz' % step,
S_step_sparse)
scipy.sparse.save_npz(params['res_dir'] + '/E_ks_%d.npz' % step,
E_step_sparse)
print(
'Done with computing molecular orbital overlaps. Elapsed time:',
time.time() - t1)
print('Done with step %d.' % step, 'Elapsed time:',
time.time() - t1_all)
else:
# The same procedure as above but now we compute time-overlaps as well
t1 = time.time()
print('Creating shell...')
shell_2, l_vals = molden_methods.molden_file_to_libint_shell('Diag_%d-libra-1_0.molden'%step,\
is_spherical)
print('Done with creating shell. Elapsed time:', time.time() - t1)
t1 = time.time()
print('Reading energies and eigenvectors....')
eig_vect_2, energies_2 = molden_methods.eigenvectors_molden('Diag_%d-libra-1_0.molden'%step,\
nbasis(shell_2),l_vals)
print('Done with reading energies and eigenvectors. Elapsed time:',
time.time() - t1)
if params['remove_molden']:
os.system('rm Diag_%d-libra-1_0.molden' % step)
print('Computing atomic orbital overlap matrix...')
t1 = time.time()
AO_S = compute_overlaps(shell_2, shell_2, nprocs)
AO_St = compute_overlaps(shell_1, shell_2, nprocs)
print('Done with computing atomic orbital overlaps. Elapsed time:',
time.time() - t1)
t1 = time.time()
print('Turning the MATRIX to numpy array...')
AO_S = data_conv.MATRIX2nparray(AO_S)
AO_St = data_conv.MATRIX2nparray(AO_St)
print('Done with transforming MATRIX 2 numpy array. Elapsed time:',
time.time() - t1)
## Now, we need to resort the eigenvectors based on the new indices
print('Resorting eigenvectors elements...')
t1 = time.time()
new_indices = CP2K_methods.resort_molog_eigenvectors(l_vals)
eigenvectors_2 = []
for j in range(len(eig_vect_2)):
# the new and sorted eigenvector
eigenvector_2 = eig_vect_2[j]
eigenvector_2 = eigenvector_2[new_indices]
# append it to the eigenvectors list
eigenvectors_2.append(eigenvector_2)
eigenvectors_2 = np.array(eigenvectors_2)
print('Done with resorting eigenvectors elements. Elapsed time:',
time.time() - t1)
##
t1 = time.time()
print('Computing and saving molecular orbital overlaps...')
S = np.linalg.multi_dot([eigenvectors_2, AO_S,
eigenvectors_2.T])[istate - 1:fstate - 1,
istate - 1:fstate - 1]
St = np.linalg.multi_dot([eigenvectors_1, AO_S,
eigenvectors_2.T])[istate - 1:fstate - 1,
istate - 1:fstate - 1]
S_step = data_conv.form_block_matrix(S, zero_mat, zero_mat, S)
St_step = data_conv.form_block_matrix(St, zero_mat, zero_mat, St)
S_step_sparse = scipy.sparse.csc_matrix(S_step)
St_step_sparse = scipy.sparse.csc_matrix(St_step)
E_step = data_conv.form_block_matrix(np.diag(energies_2)[istate-1:fstate-1,istate-1:fstate-1],\
zero_mat,zero_mat,\
np.diag(energies_2)[istate-1:fstate-1,istate-1:fstate-1])
E_step_sparse = scipy.sparse.csc_matrix(E_step)
scipy.sparse.save_npz(params['res_dir'] + '/S_ks_%d.npz' % step,
S_step_sparse)
scipy.sparse.save_npz(
params['res_dir'] + '/St_ks_%d.npz' % (step - 1),
St_step_sparse)
scipy.sparse.save_npz(params['res_dir'] + '/E_ks_%d.npz' % step,
E_step_sparse)
print(
'Done with computing molecular orbital overlaps. Elapsed time:',
time.time() - t1)
shell_1 = shell_2
energies_1 = energies_2
eigenvectors_1 = eigenvectors_2
print('Removing unnecessary wfn files...')
os.system('rm OT_%d-RESTART*' % (step - 1))
os.system('rm Diag_%d-RESTART*' % (step - 1))
print('Done with step %d.' % step, 'Elapsed time:',
time.time() - t1_all)
counter += 1
# Finally move all the pdos and log files to all_pdosfiles and all_logfiles
os.system('mv *pdos %s/.' % params['all_pdosfiles'])
os.system('mv *log %s/.' % params['all_logfiles'])
print('Done with the job!!!')
clrs_index = ["11", "21", "31", "41", "12", "22", "32", "13", "23", "14", "24"]
E_f = 2.1608
spd = ["s", "p", "d"]
all_atoms = list(range(1, 67))
Cs = [spd, all_atoms, ["Cs"]]
Sn = [spd, all_atoms, ["Sn"]]
Br = [spd, all_atoms, ["Br"]]
projections = [Cs, Sn, Br]
E, pdosa, pdosb = pdos.QE_pdos("pdos/x.pdos_atm#", -10.0, 10.0, 0.1, projections,\
E_f, "pdos_", 1, 0.01, 0.1, nspin=1)
e_grid = data_conv.MATRIX2nparray(E)
proja = data_conv.MATRIX2nparray(pdosa)
projb = data_conv.MATRIX2nparray(pdosb)
print(e_grid.shape, proja.shape, projb.shape)
def plot(_energy, _pdosa, _pdosb):
plt.rc('axes', titlesize=18) # fontsize of the axes title
plt.rc('axes', labelsize=18) # fontsize of the x and y labels
plt.rc('legend', fontsize=12) # legend fontsize
plt.rc('xtick', labelsize=18) # fontsize of the tick labels
plt.rc('ytick', labelsize=18) # fontsize of the tick labels
plt.ylim(-5.0, 5.0)
from libra_py import CP2K_methods
from libra_py import molden_methods
from libra_py import data_conv
# number of processors
nprocs = 4
# set up the timer
t1 = time.time()
# creating the shell for the molden file
shells_1, l_vals = molden_methods.molden_file_to_libint_shell('CdSe13-CdSe13-1_0.molden',True)
# extracting the eigenvectors and energies from molden file
eig1, ener1 = molden_methods.eigenvectors_molden('CdSe13-CdSe13-1_0.molden',nbasis(shells_1),l_vals)
# compute the AO overlap matrix
AO = compute_overlaps(shells_1,shells_1,nprocs)
# turn it into a numpy array
AO = data_conv.MATRIX2nparray(AO)
# new indices for the the MOLog eigenvectors
new_indices = CP2K_methods.resort_molog_eigenvectors(l_vals)
# making all the reindexed eigenvectors
eigenvectors = []
for i in range(len(eig1)):
# the new and sorted eigenvector
eigenvector = eig1[i]
eigenvector = eigenvector[new_indices]
# append it to the eigenvectors list
eigenvectors.append(eigenvector)
# make it a numpy array to be able to work with
eigs = np.array(eigenvectors)
# compute the MO overlap
S = np.linalg.multi_dot([eigs, AO ,eigs.T])
# print out the diagonal element of the MO overlap matrix
ANN.init_weights_biases_uniform(rnd, -0.1, 0.1, -0.1, 0.1)
params = {
"num_epochs": 100,
"steps_per_epoch": 10000,
"epoch_size": 32,
"learning_rate": 0.05,
"verbosity": 1
}
ANN.train(rnd, params, inputs1, targets1)
Y_train = ANN.propagate(inputs1)
Y_test = ANN.propagate(inputs2)
#sys.exit(0)
#x_train = data_conv.MATRIX2nparray(inputs1)[0][:]
#x_test = data_conv.MATRIX2nparray(inputs2)[0][:]
y_train = data_conv.MATRIX2nparray(Y_train[2])[0][:]
y_test = data_conv.MATRIX2nparray(Y_test[2])[0][:]
eg_target1 = data_conv.MATRIX2nparray(targets1)[0][:]
eg_target2 = data_conv.MATRIX2nparray(targets2)[0][:]
time1 = np.array(time1)
time2 = np.array(time2)
#print(x.shape, y.shape, eg_target2.shape)
#sys.exit(0)
plt.figure(num=None,
figsize=(3.21, 2.41),
dpi=300,
edgecolor='black',
frameon=True)
plt.subplot(1, 1, 1)
plt.title('Lumo Energy - Training',
def myfunc_st( step ):
st_sd = step3.mapping.ovlp_mat_arb( sd_states_reindexed_sorted[step], sd_states_reindexed_sorted[step+1], St_ks_job[0][step], use_minimal=False )
#print(step, "st_sd as CMATRIX <1|1> = ", st_sd.get(1,1) )
st_sd = data_conv.MATRIX2nparray(st_sd)
#print(step, "st_sd as np.array <1|1> = ", st_sd[1,1])
return st_sd
def myfunc(subtraj):
start_time = subtraj * subtraj_increment
print("\nsubtraj ", subtraj)
print("start time = ", start_time)
print("end time = ", start_time + subtraj_len)
md_time = [i for i in range(subtraj_len)]
md_time = np.array(md_time) * params["dt"] * units.au2fs
# Obtain the subtrajectory
for i in range(start_time, start_time + subtraj_len):
hvib_mb_subtrajs[subtraj].append(hvib_mb[0][i])
hvib_sd_subtrajs[subtraj].append(hvib_sd[0][i])
# Compute mb energy gaps and decoherence times
params["nsteps"] = subtraj_len
params["init_times"] = [0]
tau_mb, rates_mb = decoherence_times.decoherence_times(
hvib_mb_subtrajs[subtraj])
dE_mb = decoherence_times.energy_gaps(hvib_mb_subtrajs[subtraj])
avg_deco_mb = tau_mb * units.au2fs
print("Finished gather data for MD time",
params["init_times"][0] + params["nsteps"], "steps.")
print("Dephasing time between mb states 0 and 1 is:",
tau_mb.get(0, 1) * units.au2fs, "fs")
# Compute sd energy gaps and decoherence times
params["nsteps"] = subtraj_len
params["init_times"] = [0]
tau_sd, rates_sd = decoherence_times.decoherence_times(
hvib_sd_subtrajs[subtraj])
dE_sd = decoherence_times.energy_gaps(hvib_sd_subtrajs[subtraj])
avg_deco_sd = tau_sd * units.au2fs
print("Finished gather data for MD time",
params["init_times"][0] + params["nsteps"], "steps.")
print("Dephasing time between sd states 0 and 1 is:",
tau_sd.get(0, 1) * units.au2fs, "fs")
# Compute mb energies
mb_subtraj_energy = []
params["nstates"] = 31
for mb_index in range(params["nstates"]):
mb_subtraj_energy.append([])
for step in range(params["nsteps"]):
mb_subtraj_energy[mb_index].append(
hvib_mb_subtrajs[subtraj][step].get(mb_index, mb_index).real -
hvib_mb_subtrajs[subtraj][step].get(0, 0).real)
mb_subtraj_energy = np.array(mb_subtraj_energy)
tmp_mb_energies.append(mb_subtraj_energy)
# Compute sd energies
params["nstates"] = 64
sd_subtraj_energy = []
for sd_index in range(params["nstates"]):
sd_subtraj_energy.append([])
for step in range(params["nsteps"]):
sd_subtraj_energy[sd_index].append(
hvib_sd_subtrajs[subtraj][step].get(sd_index, sd_index).real -
hvib_sd_subtrajs[subtraj][step].get(0, 0).real)
sd_subtraj_energy = np.array(sd_subtraj_energy)
tmp_sd_energies.append(sd_subtraj_energy)
###############################################
###############################################
# Now it is time to run the NAMD
params["T"] = 300.0
params["ntraj"] = 250
params["sh_method"] = 1
params["Boltz_opt"] = 1
params["nsteps"] = subtraj_len
params["init_times"] = [0]
istates = [14, 18]
for istate in istates:
params["istate"] = istate # Recall index from 0
### FSSH
params["decoherence_method"] = 0
start = time.time()
# mb fssh
params["outfile"] = "_out_mb_fssh_subtraj_" + str(
subtraj) + "_istate" + str(istate) + ".txt"
params["nstates"] = 31 # total number of mb electronic states
res_mb_fssh = step4.run([hvib_mb_subtrajs[subtraj]], params)
mb_nbra_namd = data_conv.MATRIX2nparray(res_mb_fssh)
tmp_fssh_mb.append(mb_nbra_namd)
mb_hot_energy_fssh = (mb_nbra_namd[:, 95] -
mb_nbra_namd[:, 1]) * units.au2ev
#sys.exit(0)
#sd fssh
params["outfile"] = "_out_sd_fssh_subtraj_" + str(
subtraj) + "_istate" + str(istate) + ".txt"
params["nstates"] = 64 # total number of sd electronic states
res_sd_fssh = step4.run([hvib_sd_subtrajs[subtraj]], params)
sd_nbra_namd = data_conv.MATRIX2nparray(res_sd_fssh)
tmp_fssh_sd.append(sd_nbra_namd)
sd_hot_energy_fssh = (sd_nbra_namd[:, 194] -
sd_nbra_namd[:, 1]) * units.au2ev
end = time.time()
print("Time to run NBRA NAMD = ", end - start)
### IDA
params["decoherence_method"] = 1
start = time.time()
# mb ida
params["outfile"] = "_out_mb_ida_subtraj_" + str(
subtraj) + "_istate" + str(istate) + ".txt"
params["nstates"] = 31 # total number of mb electronic states
res_mb_ida = step4.run([hvib_mb_subtrajs[subtraj]], params)
mb_nbra_namd = data_conv.MATRIX2nparray(res_mb_ida)
tmp_ida_mb.append(mb_nbra_namd)
mb_hot_energy_ida = (mb_nbra_namd[:, 95] -
mb_nbra_namd[:, 1]) * units.au2ev
#sd ida
params["outfile"] = "_out_sd_ida_subtraj_" + str(
subtraj) + "_istate" + str(istate) + ".txt"
params["nstates"] = 64 # total number of sd electronic states
res_sd_ida = step4.run([hvib_sd_subtrajs[subtraj]], params)
sd_nbra_namd = data_conv.MATRIX2nparray(res_sd_ida)
tmp_ida_sd.append(sd_nbra_namd)
sd_hot_energy_ida = (sd_nbra_namd[:, 194] -
sd_nbra_namd[:, 1]) * units.au2ev
end = time.time()
print("Time to run NBRA NAMD = ", end - start)
### mSDM
params["decoherence_method"] = 2
start = time.time()
# mb msdm
params["outfile"] = "_out_mb_msdm_subtraj_" + str(
subtraj) + "_istate" + str(istate) + ".txt"
params["nstates"] = 31 # total number of mb electronic states
res_mb_msdm = step4.run([hvib_mb_subtrajs[subtraj]], params)
mb_nbra_namd = data_conv.MATRIX2nparray(res_mb_msdm)
tmp_msdm_mb.append(mb_nbra_namd)
mb_hot_energy_msdm = (mb_nbra_namd[:, 95] -
mb_nbra_namd[:, 1]) * units.au2ev
#sd msdm
params["outfile"] = "_out_sd_msdm_subtraj_" + str(
subtraj) + "_istate" + str(istate) + ".txt"
params["nstates"] = 64 # total number of sd electronic states
res_sd_msdm = step4.run([hvib_sd_subtrajs[subtraj]], params)
sd_nbra_namd = data_conv.MATRIX2nparray(res_sd_msdm)
tmp_msdm_sd.append(sd_nbra_namd)
sd_hot_energy_msdm = (sd_nbra_namd[:, 194] -
sd_nbra_namd[:, 1]) * units.au2ev
end = time.time()
print("Time to run NBRA NAMD = ", end - start)
plt.figure(num=None,
figsize=(3.21, 2.41),
dpi=300,
edgecolor='black',
frameon=True)
plt.subplot(1, 1, 1)
plt.title('(CdSe)$_{33}$ MB traj' + str(subtraj) + ' istate' +
str(istate) + '',
fontsize=10)
plt.xlabel('Time, fs', fontsize=10)
plt.ylabel('Energy, eV', fontsize=10)
plt.ylim(0, 3.5)
plt.yticks([0, 1, 2, 3])
for sd_index in range(31):
plt.plot(md_time,
mb_subtraj_energy[sd_index] * units.au2ev,
label="",
linewidth=1,
color="gray")
plt.plot(md_time,
mb_hot_energy_fssh,
linewidth=2.5,
color="red",
label="FSSH")
plt.plot(md_time,
mb_hot_energy_ida,
linewidth=2.5,
color="green",
label="IDA")
plt.plot(md_time,
mb_hot_energy_msdm,
linewidth=2.5,
color="blue",
label="mSDM")
plt.tight_layout()
plt.legend(fontsize=8, ncol=3, loc="lower left")
plt.savefig("CdSe33_MB_Dyn_" + str(subtraj) + "_" + str(istate) +
".png",
dpi=300)
plt.figure(num=None,
figsize=(3.21, 2.41),
dpi=300,
edgecolor='black',
frameon=True)
plt.subplot(1, 1, 1)
plt.title('(CdSe)$_{33}$ SP traj' + str(subtraj) + ' istate' +
str(istate) + '',
fontsize=10)
plt.xlabel('Time, fs', fontsize=10)
plt.ylabel('Energy, eV', fontsize=10)
plt.ylim(0, 3.5)
plt.yticks([0, 1, 2, 3])
for sd_index in range(64):
plt.plot(md_time,
sd_subtraj_energy[sd_index] * units.au2ev,
label="",
linewidth=1,
color="gray")
plt.plot(md_time,
sd_hot_energy_fssh,
linewidth=2.5,
color="red",
label="FSSH")
plt.plot(md_time,
sd_hot_energy_ida,
linewidth=2.5,
color="green",
label="IDA")
plt.plot(md_time,
sd_hot_energy_msdm,
linewidth=2.5,
color="blue",
label="mSDM")
plt.tight_layout()
plt.legend(fontsize=8, ncol=3, loc="lower left")
plt.savefig("CdSe33_" + str(subtraj) + "_SP_Dyn_" + str(istate) +
".png",
dpi=300)
def plot_training(_plt,
_ann,
_training_input,
_training_target,
mode1=0,
mode2=2,
filename="ANN_training.png"):
"""
Args:
_plt (matplotlib instance)
_ann (NeuralNetwork object)
_training_input ( MATRIX(n_inputs, n_patterns )
_training_target ( MATRIX(n_outputs, n_patterns )
"""
last_indx = _ann.Nlayers - 1
# Predict
# There is only 1 output in this case - we take the first index
training_out = _ann.propagate(_training_input)
y_predict = data_conv.MATRIX2nparray(training_out[last_indx])[0, :]
#print(F"len(training_out) = {len(training_out)}")
#print(F"len(y_predict) = {len(y_predict)}")
# Given targets
# Targets also have only 1 components in this case
y_targets = data_conv.MATRIX2nparray(_training_target)[0, :]
#print(F"len(y_targets) = {len(y_targets)}")
nsteps = len(y_predict) # one less, so it would work for NACs too
steps_train = list(range(nsteps))
#print(F"nsteps = {nsteps}")
#print(F"steps_trian = {steps_train}")
x_mode1 = data_conv.MATRIX2nparray(_training_input)[mode1, :]
x_mode2 = data_conv.MATRIX2nparray(_training_input)[mode2, :]
#print(x_mode1)
#print(x_mode2)
_plt.rc('axes', titlesize=12) # fontsize of the axes title
_plt.rc('axes', labelsize=12) # fontsize of the x and y labels
_plt.rc('legend', fontsize=10) # legend fontsize
_plt.rc('xtick', labelsize=8) # fontsize of the tick labels
_plt.rc('ytick', labelsize=8) # fontsize of the tick labels
_plt.rc('figure.subplot', left=0.2)
_plt.rc('figure.subplot', right=0.95)
_plt.rc('figure.subplot', bottom=0.13)
_plt.rc('figure.subplot', top=0.88)
figure = _plt.figure(num=None,
figsize=(3.21 * 3, 2.41),
dpi=300,
edgecolor='black',
frameon=True)
_plt.subplot(1, 3, 1)
_plt.xlabel('Timestep', fontsize=10)
_plt.ylabel('Property', fontsize=10)
_plt.plot(steps_train,
y_predict[:nsteps],
color="black",
label="ANN",
linewidth=2)
_plt.plot(steps_train,
y_targets[:nsteps],
color="green",
label="reference",
linewidth=2)
_plt.legend(fontsize=6.75, ncol=1, loc='upper left')
_plt.subplot(1, 3, 2)
_plt.xlabel(F'Mode {mode1}', fontsize=10)
_plt.ylabel('Property', fontsize=10)
x_mode1 = data_conv.MATRIX2nparray(_training_input)[mode1, :]
_plt.plot(x_mode1[:nsteps],
y_predict[:nsteps],
color="black",
label="ANN",
linewidth=2)
_plt.plot(x_mode1[:nsteps],
y_targets[:nsteps],
color="green",
label="reference",
linewidth=2)
_plt.legend(fontsize=6.75, ncol=1, loc='upper left')
_plt.subplot(1, 3, 3)
_plt.xlabel(F'Mode {mode2}', fontsize=10)
_plt.ylabel('Property', fontsize=10)
x_mode2 = data_conv.MATRIX2nparray(_training_input)[mode2, :]
_plt.plot(x_mode2[:nsteps],
y_predict[:nsteps],
color="black",
label="ANN",
linewidth=2)
_plt.plot(x_mode2[:nsteps],
y_targets[:nsteps],
color="green",
label="reference",
linewidth=2)
_plt.legend(fontsize=6.75, ncol=1, loc='upper left')
_plt.tight_layout()
_plt.savefig(filename, dpi=300)
_plt.show()
