def run(self, r_data=None):
"""
Once all attributes are setup as you so desire, run this method to
optimize the parameters.
Returns
-------
`datatypes.FF` (or subclass)
Contains the best parameters.
"""
if r_data is None:
r_data = opt.return_ref_data(self.args_ref)
if self.ff.score is None:
logger.log(20, '~~ CALCULATING INITIAL FF SCORE ~~'.rjust(79, '~'))
self.ff.export_ff()
# Could store data on self.ff.data if we wanted. Not necessary for
# simplex. If simplex yielded no improvements, it would return this
# FF, and then we might want the data such taht we don't have to
# recalculate it in gradient. Let's hope simplex generally yields
# improvements.
data = calculate.main(self.args_ff)
self.ff.score = compare.compare_data(r_data, data)
else:
logger.log(20, ' -- Reused existing score and data for initial FF.')
logger.log(20, '~~ SIMPLEX OPTIMIZATION ~~'.rjust(79, '~'))
logger.log(20, 'INIT FF SCORE: {}'.format(self.ff.score))
opt.pretty_ff_results(self.ff, level=20)
if self.max_params and len(self.ff.params) > self.max_params:
logger.log(20, ' -- More parameters than the maximum allowed.')
logger.log(5, 'CURRENT PARAMS: {}'.format(len(self.ff.params)))
logger.log(5, 'MAX PARAMS: {}'.format(self.max_params))
# Here we select the parameters that have the lowest 2nd
# derivatives.
# THIS IS SCHEDULED FOR CHANGING. THIS IS ACTUALLY NOT A GOOD
# CRITERION FOR PARAMETER SELECTION.
if self.ff.params[0].d1:
logger.log(15, ' -- Reusing existing parameter derivatives.')
# Differentiate all parameters forward. Yes, I know this is
# counter-intuitive because we are going to only use subset of
# the forward differentiated FFs. However, this is very
# computationally inexpensive because we're not scoring them
# now. We will remove the forward differentiated FFs we don't
# want before scoring.
ffs = opt.differentiate_ff(self.ff, central=False)
else:
logger.log(15, ' -- Calculating new parameter derivatives.')
# Do central differentiation so we can calculate derivatives.
# Another option would be to write code to determine
# derivatives only from forward differentiation.
ffs = opt.differentiate_ff(self.ff, central=True)
# We have to score to get the derivatives.
for ff in ffs:
ff.export_ff(lines=self.ff_lines)
logger.log(20, ' -- Calculating {}.'.format(ff))
data = calculate.main(self.args_ff)
ff.score = compare.compare_data(r_data, data)
opt.pretty_ff_results(ff)
# Add the derivatives to your original FF.
opt.param_derivs(self.ff, ffs)
# Only keep the forward differentiated FFs.
ffs = opt.extract_forward(ffs)
logger.log(5, ' -- Keeping {} forward differentiated '
'FFs.'.format(len(ffs)))
# This sorts the parameters based upon their 2nd derivative.
# It keeps the ones with lowest 2nd derivatives.
# SCHEDULED FOR CHANGES. NOT A GOOD SORTING CRITERION.
params = select_simp_params_on_derivs(
self.ff.params, max_params=self.max_params)
# From the entire list of forward differentiated FFs, pick
# out the ones that have the lowest 2nd derivatives.
self.new_ffs = opt.extract_ff_by_params(ffs, params)
logger.log(1, '>>> len(self.new_ffs): {}'.format(len(self.new_ffs)))
# Reduce number of parameters.
# Will need an option that's not MM3* specific in the future.
ff_rows = [x.mm3_row for x in params]
ff_cols = [x.mm3_col for x in params]
for ff in self.new_ffs:
new_params = []
for param in ff.params:
if param.mm3_row in ff_rows and param.mm3_col in ff_cols:
new_params.append(param)
ff.params = new_params
# Make a copy of your original FF that has less parameters.
ff_copy = copy.deepcopy(self.ff)
new_params = []
for param in ff.params:
if param.mm3_row in ff_rows and param.mm3_col in ff_cols:
new_params.append(param)
ff_copy.params = new_params
else:
# In this case it's simple. Just forward differentiate each
# parameter.
self.new_ffs = opt.differentiate_ff(self.ff, central=False)
logger.log(1, '>>> len(self.new_ffs): {}'.format(len(self.new_ffs)))
# Still make that FF copy.
ff_copy = copy.deepcopy(self.ff)
# Double check and make sure they're all scored.
for ff in self.new_ffs:
if ff.score is None:
ff.export_ff(lines=self.ff_lines)
logger.log(20, ' -- Calculating {}.'.format(ff))
data = calculate.main(self.args_ff)
ff.score = compare.compare_data(r_data, data)
opt.pretty_ff_results(ff)
# Add your copy of the orignal to FF to the forward differentiated FFs.
self.new_ffs = sorted(self.new_ffs + [ff_copy], key=lambda x: x.score)
# Allow 3 cycles w/o change for each parameter present. Remember that
# the initial FF was added here, hence the minus one.
self._max_cycles_wo_change = 3 * (len(self.new_ffs) - 1)
wrapper = textwrap.TextWrapper(width=79)
# Shows all FFs parameters.
opt.pretty_ff_params(self.new_ffs)
# Start the simplex cycles.
current_cycle = 0
cycles_wo_change = 0
while current_cycle < self.max_cycles \
and cycles_wo_change < self._max_cycles_wo_change:
current_cycle += 1
last_best = self.new_ffs[0].score
best_ff = self.new_ffs[0]
logger.log(20, '~~ START SIMPLEX CYCLE {} ~~'.format(
current_cycle).rjust(79, '~'))
logger.log(20, 'ORDERED FF SCORES:')
logger.log(20, wrapper.fill('{}'.format(
' '.join('{:15.4f}'.format(x.score) for x in self.new_ffs))))
# !!! FOR TESTING !!!
# Write the best and worst FFs to some other directory. Then
# write the worst FF to optimization working directory. Then
# raise opt.OptError. The worst FF should be overwritten by
# the best FF afterwards.
# if current_cycle == 5:
# self.new_ffs[-1].export_ff(
# path='ref_methanol_flds/mm3_worst.fld',
# lines=self.ff.lines)
# self.new_ffs[0].export_ff(
# path='ref_methanol_flds/mm3_best.fld',
# lines=self.ff.lines)
# self.new_ffs[-1].export_ff(
# path='ref_methanol/mm3.fld',
# lines=self.ff.lines)
# raise opt.OptError
# !!! END TESTING !!!
inv_ff = self.ff.__class__()
if self.do_weighted_reflection:
inv_ff.method = 'WEIGHTED INVERSION'
else:
inv_ff.method = 'INVERSION'
inv_ff.params = copy.deepcopy(best_ff.params)
ref_ff = self.ff.__class__()
ref_ff.method = 'REFLECTION'
ref_ff.params = copy.deepcopy(best_ff.params)
# Need score difference sum for weighted inversion.
# Calculate this value before going into loop.
if self.do_weighted_reflection:
# If zero, should break.
score_diff_sum = sum([x.score - self.new_ffs[-1].score
for x in self.new_ffs[:-1]])
if score_diff_sum == 0.:
logger.warning(
'No difference between force field scores. '
'Exiting simplex.')
# We want to raise opt.OptError such that
# opt.catch_run_errors will write the best FF obtained thus
# far.
raise opt.OptError(
'No difference between force field scores. '
'Exiting simplex.')
for i in xrange(0, len(best_ff.params)):
if self.do_weighted_reflection:
inv_val = (
sum([x.params[i].value *
(x.score - self.new_ffs[-1].score)
for x in self.new_ffs[:-1]])
/ score_diff_sum)
else:
inv_val = (
sum([x.params[i].value for x in self.new_ffs[:-1]])
/
len(self.new_ffs[:-1]))
inv_ff.params[i].value = inv_val
ref_ff.params[i].value = (
2 * inv_val - self.new_ffs[-1].params[i].value)
# The inversion point does not need to be scored.
# Calculate score for reflected parameters.
self.ff.export_ff(self.ff.path, params=ref_ff.params)
data = calculate.main(self.args_ff)
ref_ff.score = compare.compare_data(r_data, data)
opt.pretty_ff_results(ref_ff)
if ref_ff.score < self.new_ffs[0].score:
logger.log(20, '~~ ATTEMPTING EXPANSION ~~'.rjust(79, '~'))
exp_ff = self.ff.__class__()
exp_ff.method = 'EXPANSION'
exp_ff.params = copy.deepcopy(best_ff.params)
for i in xrange(0, len(self.new_ffs[0].params)):
exp_ff.params[i].value = (
3 * inv_ff.params[i].value -
2 * self.new_ffs[-1].params[i].value)
self.ff.export_ff(self.ff.path, exp_ff.params)
data = calculate.main(self.args_ff)
exp_ff.score = compare.compare_data(r_data, data)
opt.pretty_ff_results(exp_ff)
if exp_ff.score < ref_ff.score:
self.new_ffs[-1] = exp_ff
logger.log(
20, ' -- Expansion succeeded. Keeping expanded '
'parameters.')
else:
self.new_ffs[-1] = ref_ff
logger.log(
20, ' -- Expansion failed. Keeping reflected parameters.')
elif ref_ff.score < self.new_ffs[-2].score:
logger.log(20, ' -- Keeping reflected parameters.')
self.new_ffs[-1] = ref_ff
else:
logger.log(20, '~~ ATTEMPTING CONTRACTION ~~'.rjust(79, '~'))
con_ff = self.ff.__class__()
con_ff.method = 'CONTRACTION'
con_ff.params = copy.deepcopy(best_ff.params)
for i in xrange(0, len(best_ff.params)):
if ref_ff.score > self.new_ffs[-1].score:
con_val = (
(inv_ff.params[i].value +
self.new_ffs[-1].params[i].value) / 2)
else:
con_val = (
(3 * inv_ff.params[i].value -
self.new_ffs[-1].params[i].value) / 2)
con_ff.params[i].value = con_val
self.ff.export_ff(self.ff.path, params=con_ff.params)
data = calculate.main(self.args_ff)
con_ff.score = compare.compare_data(r_data, data)
opt.pretty_ff_results(con_ff)
# This change was made to reflect the 1998 Q2MM publication.
# if con_ff.score < self.new_ffs[-1].score:
if con_ff.score < self.new_ffs[-2].score:
logger.log(20, ' -- Contraction succeeded.')
self.new_ffs[-1] = con_ff
elif self.do_massive_contraction:
logger.log(
20, '~~ DOING MASSIVE CONTRACTION ~~'.rjust(79, '~'))
for ff_num, ff in enumerate(self.new_ffs[1:]):
for i in xrange(0, len(best_ff.params)):
ff.params[i].value = (
(ff.params[i].value +
self.new_ffs[0].params[i].value) / 2)
self.ff.export_ff(self.ff.path, params=ff.params)
data = calculate.main(self.args_ff)
ff.score = compare.compare_data(r_data, data)
ff.method += ' MC'
opt.pretty_ff_results(ff)
else:
logger.log(
20, ' -- Contraction failed. Keeping parmaeters '
'anyway.')
self.new_ffs[-1] = con_ff
self.new_ffs = sorted(self.new_ffs, key=lambda x: x.score)
if self.new_ffs[0].score < last_best:
cycles_wo_change = 0
else:
cycles_wo_change += 1
logger.log(20, ' -- {} cycles without improvement out of {} '
'allowed.'.format(
cycles_wo_change, self._max_cycles_wo_change))
best_ff = self.new_ffs[0]
logger.log(20, 'BEST:')
opt.pretty_ff_results(self.new_ffs[0], level=20)
logger.log(20, '~~ END SIMPLEX CYCLE {} ~~'.format(
current_cycle).rjust(79, '~'))
if best_ff.score < self.ff.score:
logger.log(20, '~~ SIMPLEX FINISHED WITH IMPROVEMENTS ~~'.rjust(
79, '~'))
best_ff = restore_simp_ff(best_ff, self.ff)
else:
logger.log(20, '~~ SIMPLEX FINISHED WITHOUT IMPROVEMENTS ~~'.rjust(
79, '~'))
opt.pretty_ff_results(self.ff, level=20)
opt.pretty_ff_results(best_ff, level=20)
logger.log(20, ' -- Writing best force field from simplex.')
best_ff.export_ff(best_ff.path)
return best_ff
def run(self, ref_data=None, restart=None):
"""
Runs the gradient optimization.
Ensure that the attributes in __init__ are set as you desire before
using this function.
Returns
-------
`datatypes.FF` (or subclass)
Contains the best parameters.
"""
# We need reference data if you didn't provide it.
if ref_data is None:
ref_data = opt.return_ref_data(self.args_ref)
# We need the initial FF data.
if self.ff.data is None:
logger.log(20, '~~ GATHERING INITIAL FF DATA ~~'.rjust(79, '~'))
# Is opt.Optimizer.ff_lines used anymore?
self.ff.export_ff()
self.ff.data = calculate.main(self.args_ff)
# Not 100% sure if this is necessary, but it certainly doesn't hurt.
compare.correlate_energies(ref_data, self.ff.data)
r_dict = compare.data_by_type(ref_data)
c_dict = compare.data_by_type(self.ff.data)
r_dict, c_dict = compare.trim_data(r_dict, c_dict)
if self.ff.score is None:
# Already zeroed reference and correlated the energies.
self.ff.score = compare.compare_data(r_dict, c_dict)
data_types = []
for typ in r_dict:
data_types.append(typ)
data_types.sort()
logger.log(20, '~~ GRADIENT OPTIMIZATION ~~'.rjust(79, '~'))
logger.log(20, 'INIT FF SCORE: {}'.format(self.ff.score))
opt.pretty_ff_results(self.ff, level=20)
logger.log(20, '~~ CENTRAL DIFFERENTIATION ~~'.rjust(79, '~'))
if restart:
par_file = restart
logger.log(
20, ' -- Restarting gradient from central '
'differentiation file {}.'.format(par_file))
else:
# We need a file to hold the differentiated parameter data.
par_files = glob.glob(os.path.join(self.direc, 'par_diff_???.txt'))
if par_files:
par_files.sort()
most_recent_par_file = par_files[-1]
most_recent_par_file = most_recent_par_file.split('/')[-1]
most_recent_num = most_recent_par_file[9:12]
num = int(most_recent_num) + 1
par_file = 'par_diff_{:03d}.txt'.format(num)
else:
par_file = 'par_diff_001.txt'
logger.log(
20, ' -- Generating central differentiation '
'file {}.'.format(par_file))
f = open(os.path.join(self.direc, par_file), 'w')
csv_writer = csv.writer(f)
# Row 1 - Labels
# Row 2 - Weights
# Row 3 - Reference data values
# Row 4 - Initial FF data values
## Deprecated -TR
#csv_writer.writerow([x.lbl for x in ref_data])
#csv_writer.writerow([x.wht for x in ref_data])
#csv_writer.writerow([x.val for x in ref_data])
#csv_writer.writerow([x.val for x in self.ff.data])
writerows = [[], [], [], []]
for data_type in data_types:
writerows[0].extend([x.lbl for x in r_dict[data_type]])
writerows[1].extend([x.wht for x in r_dict[data_type]])
writerows[2].extend([x.val for x in r_dict[data_type]])
writerows[3].extend([x.val for x in c_dict[data_type]])
for row in writerows:
csv_writer.writerow(row)
logger.log(20, '~~ DIFFERENTIATING PARAMETERS ~~'.rjust(79, '~'))
# Save many FFs, each with their own parameter sets.
ffs = opt.differentiate_ff(self.ff)
logger.log(
20, '~~ SCORING DIFFERENTIATED PARAMETERS ~~'.rjust(79, '~'))
for ff in ffs:
ff.export_ff(lines=self.ff.lines)
logger.log(20, ' -- Calculating {}.'.format(ff))
data = calculate.main(self.args_ff)
# Deprecated
#ff.score = compare.compare_data(ref_data, data)
c_data = compare.data_by_type(data)
r_dict, c_data = compare.trim_data(r_dict, c_data)
ff.score = compare.compare_data(r_dict, c_data)
opt.pretty_ff_results(ff)
# Write the data rather than storing it in memory. For large
# parameter sets, this could consume GBs of memory otherwise!
#csv_writer.writerow([x.val for x in data])
row = []
for data_type in data_types:
row.extend([x.val for x in c_data[data_type]])
csv_writer.writerow(row)
f.close()
# Make sure we have derivative information. Used for NR.
#
# The derivatives are useful for checking up on the progress of the
# optimization and for deciding which parameters to use in a
# subsequent simplex optimization.
#
# Still need a way to do this with the resatrt file.
opt.param_derivs(self.ff, ffs)
# Calculate the Jacobian, residual vector, matrix A and vector b.
# These aren't needed if you're only doing Newton-Raphson.
if self.do_lstsq or self.do_lagrange or self.do_levenberg or \
self.do_svd:
logger.log(20, '~~ JACOBIAN AND RESIDUAL VECTOR ~~'.rjust(79, '~'))
# Setup the residual vector.
# Deprecated - TR
#num_d = len(ref_data)
num_d = 0
for datatype in r_dict:
num_d += len(r_dict[datatype])
resid = np.empty((num_d, 1), dtype=float)
# Deprecated - TR
#for i in xrange(0, num_d):
# resid[i, 0] = ref_data[i].wht * \
# (ref_data[i].val - self.ff.data[i].val)
count = 0
for data_type in data_types:
for r, c in zip(r_dict[data_type], c_dict[data_type]):
resid[count, 0] = r.wht * (r.val - c.val)
count += 1
# logger.log(5, 'RESIDUAL VECTOR:\n{}'.format(resid))
logger.log(20,
' -- Formed {} residual vector.'.format(resid.shape))
# Setup the Jacobian.
num_p = len(self.ff.params)
# Maybe should be a part of the Jacobian function.
jacob = np.empty((num_d, num_p), dtype=float)
jacob = return_jacobian(jacob, os.path.join(self.direc, par_file))
# logger.log(5, 'JACOBIAN:\n{}'.format(jacob))
logger.log(20, ' -- Formed {} Jacobian.'.format(jacob.shape))
ma = jacob.T.dot(jacob)
vb = jacob.T.dot(resid)
# We need these for most optimization methods.
logger.log(5, ' MATRIX A AND VECTOR B '.center(79, '-'))
# logger.log(5, 'A:\n{}'.format(ma))
# logger.log(5, 'b:\n{}'.format(vb))
# Start coming up with new parameter sets.
if self.do_newton and not restart:
logger.log(20, '~~ NEWTON-RAPHSON ~~'.rjust(79, '~'))
# Moved the derivative section outside of here.
changes = do_newton(self.ff.params,
radii=self.newton_radii,
cutoffs=self.newton_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
if self.do_lstsq:
logger.log(20, '~~ LEAST SQUARES ~~'.rjust(79, '~'))
changes = do_lstsq(ma,
vb,
radii=self.lstsq_radii,
cutoffs=self.lstsq_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
if self.do_lagrange:
logger.log(20, '~~ LAGRANGE ~~'.rjust(79, '~'))
for factor in sorted(self.lagrange_factors):
changes = do_lagrange(ma,
vb,
factor,
radii=self.lagrange_radii,
cutoffs=self.lagrange_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
if self.do_levenberg:
logger.log(20, '~~ LEVENBERG ~~'.rjust(79, '~'))
for factor in sorted(self.levenberg_factors):
changes = do_levenberg(ma,
vb,
factor,
radii=self.levenberg_radii,
cutoffs=self.levenberg_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
if self.do_svd:
logger.log(20, '~~ SINGULAR VALUE DECOMPOSITION ~~'.rjust(79, '~'))
# J = U . s . VT
mu, vs, mvt = return_svd(jacob)
logger.log(1, '>>> mu.shape: {}'.format(mu.shape))
logger.log(1, '>>> vs.shape: {}'.format(vs.shape))
logger.log(1, '>>> mvt.shape: {}'.format(mvt.shape))
logger.log(1, '>>> vb.shape: {}'.format(vb.shape))
if self.svd_factors:
changes = do_svd_w_thresholds(mu,
vs,
mvt,
resid,
self.svd_factors,
radii=self.svd_radii,
cutoffs=self.svd_cutoffs)
else:
changes = do_svd_wo_thresholds(mu,
vs,
mvt,
resid,
radii=self.svd_radii,
cutoffs=self.svd_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
# Report how many trial FFs were generated.
logger.log(
20, ' -- Generated {} trial force field(s).'.format(
len(self.new_ffs)))
# If there are any trials, test them.
if self.new_ffs:
logger.log(20, '~~ EVALUATING TRIAL FF(S) ~~'.rjust(79, '~'))
for ff in self.new_ffs:
data = opt.cal_ff(ff, self.args_ff, parent_ff=self.ff)
# Shouldn't need to zero anymore.
# Deprecated
#ff.score = compare.compare_data(ref_data, data)
c_data = compare.data_by_type(data)
r_dict, c_data = compare.trim_data(r_dict, c_data)
ff.score = compare.compare_data(r_dict, c_data)
opt.pretty_ff_results(ff)
self.new_ffs = sorted(self.new_ffs, key=lambda x: x.score)
# Check for improvement.
if self.new_ffs[0].score < self.ff.score:
ff = self.new_ffs[0]
logger.log(
20,
'~~ GRADIENT FINISHED WITH IMPROVEMENTS ~~'.rjust(79, '~'))
opt.pretty_ff_results(self.ff, level=20)
opt.pretty_ff_results(ff, level=20)
# Copy parameter derivatives from original FF to save time in
# case we move onto simplex immediately after this.
copy_derivs(self.ff, ff)
else:
ff = self.ff
else:
ff = self.ff
ff.export_ff(ff.path)
return ff
def run(self, r_data=None):
"""
Once all attributes are setup as you so desire, run this method to
optimize the parameters.
Returns
-------
`datatypes.FF` (or subclass)
Contains the best parameters.
"""
if r_data is None:
r_data = opt.return_ref_data(self.args_ref)
logger.log(20, '~~ SIMPLEX OPTIMIZATION ~~'.rjust(79, '~'))
# Here we don't actually need the database connection/force field data.
# We only need the score.
if self.ff.score is None:
logger.log(20, '~~ CALCULATING INITIAL FF SCORE ~~'.rjust(79, '~'))
self.ff.export_ff()
# I could store this object to prevent on self.ff to prevent garbage
# collection. Would be nice if simplex was followed by gradient,
# which needs that information, and if simplex yielded no
# improvements. At most points in the optimization, this is probably
# too infrequent for it to be worth the memory, but it might be nice
# once the parameters are close to convergence.
data = calculate.main(self.args_ff)
self.ff.score = compare.compare_data(r_data, data, zero=False)
logger.log(20, 'INITIAL FF SCORE: {}'.format(self.ff.score))
else:
logger.log(15, ' -- Reused existing score and data for initial FF.')
logger.log(15, 'INIT FF SCORE: {}'.format(self.ff.score))
ffs = opt.differentiate_ff(self.ff)
for ff in ffs:
ff.export_ff(lines=self.ff_lines)
logger.log(20, ' -- Calculating {}.'.format(ff))
data = calculate.main(self.args_ff)
ff.score = compare.compare_data(r_data, data, zero=False)
opt.pretty_ff_results(ff)
if self.max_params and len(self.ff.params) > self.max_params:
simp_params = reduce_num_simp_params(
self.ff, ffs, max_params=self.max_params)
self.new_ffs = reduce_num_simp_ffs(
ffs, simp_params)
else:
self.new_ffs = ffs
self.new_ffs = sorted(self.new_ffs + [self.ff], key=lambda x: x.score)
wrapper = textwrap.TextWrapper(width=79)
logger.log(20, 'ORDERED FF SCORES:')
logger.log(20, wrapper.fill('{}'.format(
' '.join('{:15.4f}'.format(x.score) for x in self.new_ffs))))
# Shows all FFs parameters.
opt.pretty_ff_params(self.new_ffs)
# Start the simplex cycles.
current_cycle = 0
cycles_wo_change = 0
while current_cycle < self.max_cycles \
and cycles_wo_change < self.max_cycles_wo_change:
current_cycle += 1
last_best = self.new_ffs[0].score
best_ff = self.new_ffs[0]
logger.log(20, '~~ START SIMPLEX CYCLE {} ~~'.format(
current_cycle).rjust(79, '~'))
inv_ff = self.ff.__class__()
if self.do_weighted_reflection:
inv_ff.method = 'WEIGHTED INVERSION'
else:
inv_ff.method = 'INVERSION'
inv_ff.params = copy.deepcopy(best_ff.params)
ref_ff = self.ff.__class__()
ref_ff.method = 'REFLECTION'
ref_ff.params = copy.deepcopy(best_ff.params)
for i in xrange(0, len(best_ff.params)):
if self.do_weighted_reflection:
try:
inv_val = (
sum([x.params[i].value *
(x.score - self.new_ffs[-1].score)
for x in self.new_ffs[:-1]])
/
sum([x.score - self.new_ffs[-1].score
for x in self.new_ffs[:-1]]))
except ZeroDivisionError:
logger.warning(
'Attempted to divide by zero while calculating the '
'weighted simplex inversion point. All penalty '
'function scores for the trial force fields are '
'numerically equivalent.')
# Breaking should just exit the while loop. Should still
# give you the best force field determined thus far.
break
else:
inv_val = (
sum([x.params[i].value for x in self.new_ffs[:-1]])
/
len(self.new_ffs[:-1]))
inv_ff.params[i].value = inv_val
ref_ff.params[i].value = (
2 * inv_val - self.new_ffs[-1].params[i].value)
# Calculate score for inverted parameters.
self.ff.export_ff(self.ff.path, params=inv_ff.params)
data = calculate.main(self.args_ff)
inv_ff.score = compare.compare_data(r_data, data, zero=False)
opt.pretty_ff_results(inv_ff)
# Calculate score for reflected parameters.
self.ff.export_ff(self.ff.path, params=ref_ff.params)
data = calculate.main(self.args_ff)
ref_ff.score = compare.compare_data(r_data, data, zero=False)
opt.pretty_ff_results(ref_ff)
if ref_ff.score < self.new_ffs[0].score:
logger.log(20, '~~ ATTEMPTING EXPANSION ~~'.rjust(79, '~'))
exp_ff = self.ff.__class__()
exp_ff.method = 'EXPANSION'
exp_ff.params = copy.deepcopy(best_ff.params)
for i in xrange(0, len(self.new_ffs[0].params)):
exp_ff.params[i].value = (
3 * inv_ff.params[i].value -
2 * self.new_ffs[-1].params[i].value)
self.ff.export_ff(self.ff.path, exp_ff.params)
data = calculate.main(self.args_ff)
exp_ff.score = compare.compare_data(r_data, data, zero=False)
opt.pretty_ff_results(exp_ff)
if exp_ff.score < ref_ff.score:
self.new_ffs[-1] = exp_ff
logger.log(
20, ' -- Expansion succeeded. Keeping expanded '
'parameters.')
else:
self.new_ffs[-1] = ref_ff
logger.log(
20, ' -- Expansion failed. Keeping reflected parameters.')
elif ref_ff.score < self.new_ffs[-2].score:
logger.log(20, ' -- Keeping reflected parameters.')
self.new_ffs[-1] = ref_ff
else:
logger.log(20, '~~ ATTEMPTING CONTRACTION ~~'.rjust(79, '~'))
con_ff = self.ff.__class__()
con_ff.method = 'CONTRACTION'
con_ff.params = copy.deepcopy(best_ff.params)
for i in xrange(0, len(best_ff.params)):
if ref_ff.score > self.new_ffs[-1].score:
con_val = (
(inv_ff.params[i].value +
self.new_ffs[-1].params[i].value) / 2)
else:
con_val = (
(3 * inv_ff.params[i].value -
self.new_ffs[-1].params[i].value) / 2)
con_ff.params[i].value = con_val
self.ff.export_ff(self.ff.path, params=con_ff.params)
data = calculate.main(self.args_ff)
con_ff.score = compare.compare_data(r_data, data, zero=False)
opt.pretty_ff_results(con_ff)
if con_ff.score < self.new_ffs[-2].score:
self.new_ffs[-1] = con_ff
elif self.do_massive_contraction:
logger.log(
20, '~~ DOING MASSIVE CONTRACTION ~~'.rjust(79, '~'))
for ff_num, ff in enumerate(self.new_ffs[1:]):
for i in xrange(0, len(best_ff.params)):
ff.params[i].value = (
(ff.params[i].value +
self.new_ffs[0].params[i].value) / 2)
self.ff.export_ff(self.ff.path, params=ff.params)
data = calculate.main(self.args_ff)
ff.score = compare.compare_data(r_data, data, zero=False)
ff.method += ' MC'
opt.pretty_ff_results(ff)
else:
logger.log(20, ' -- Contraction failed.')
self.new_ffs = sorted(self.new_ffs, key=lambda x: x.score)
if self.new_ffs[0].score < last_best:
cycles_wo_change = 0
else:
cycles_wo_change += 1
logger.log(20, ' -- {} cycles without change.'.format(
cycles_wo_change))
best_ff = self.new_ffs[0]
logger.log(20, 'BEST:')
opt.pretty_ff_results(self.new_ffs[0], level=20)
logger.log(20, '~~ END SIMPLEX CYCLE {} ~~'.format(
current_cycle).rjust(79, '~'))
if best_ff.score < self.ff.score:
logger.log(20, '~~ SIMPLEX FINISHED WITH IMPROVEMENTS ~~'.rjust(
79, '~'))
best_ff = restore_simp_ff(best_ff, self.ff)
else:
logger.log(20, '~~ SIMPLEX FINISHED WITHOUT IMPROVEMENTS ~~'.rjust(
79, '~'))
opt.pretty_ff_results(self.ff, level=20)
opt.pretty_ff_results(best_ff, level=20)
logger.log(20, ' -- Writing best force field from simplex.')
best_ff.export_ff(best_ff.path)
return best_ff
def run(self, ref_data=None, restart=None):
"""
Runs the gradient optimization.
Ensure that the attributes in __init__ are set as you desire before
using this function.
Returns
-------
`datatypes.FF` (or subclass)
Contains the best parameters.
"""
# We need reference data if you didn't provide it.
if ref_data is None:
ref_data = opt.return_ref_data(self.args_ref)
# We need the initial FF data.
if self.ff.data is None:
logger.log(20, '~~ GATHERING INITIAL FF DATA ~~'.rjust(79, '~'))
# Is opt.Optimizer.ff_lines used anymore?
self.ff.export_ff()
self.ff.data = calculate.main(self.args_ff)
# Not 100% sure if this is necessary, but it certainly doesn't hurt.
compare.correlate_energies(ref_data, self.ff.data)
r_dict = compare.data_by_type(ref_data)
c_dict = compare.data_by_type(self.ff.data)
r_dict, c_dict = compare.trim_data(r_dict,c_dict)
if self.ff.score is None:
# Already zeroed reference and correlated the energies.
self.ff.score = compare.compare_data(r_dict, c_dict)
data_types = []
for typ in r_dict:
data_types.append(typ)
data_types.sort()
logger.log(20, '~~ GRADIENT OPTIMIZATION ~~'.rjust(79, '~'))
logger.log(20, 'INIT FF SCORE: {}'.format(self.ff.score))
opt.pretty_ff_results(self.ff, level=20)
logger.log(20, '~~ CENTRAL DIFFERENTIATION ~~'.rjust(79, '~'))
if restart:
par_file = restart
logger.log(20, ' -- Restarting gradient from central '
'differentiation file {}.'.format(par_file))
else:
# We need a file to hold the differentiated parameter data.
par_files = glob.glob(os.path.join(self.direc, 'par_diff_???.txt'))
if par_files:
par_files.sort()
most_recent_par_file = par_files[-1]
most_recent_par_file = most_recent_par_file.split('/')[-1]
most_recent_num = most_recent_par_file[9:12]
num = int(most_recent_num) + 1
par_file = 'par_diff_{:03d}.txt'.format(num)
else:
par_file = 'par_diff_001.txt'
logger.log(20, ' -- Generating central differentiation '
'file {}.'.format(par_file))
f = open(os.path.join(self.direc, par_file), 'w')
csv_writer = csv.writer(f)
# Row 1 - Labels
# Row 2 - Weights
# Row 3 - Reference data values
# Row 4 - Initial FF data values
## Deprecated -TR
#csv_writer.writerow([x.lbl for x in ref_data])
#csv_writer.writerow([x.wht for x in ref_data])
#csv_writer.writerow([x.val for x in ref_data])
#csv_writer.writerow([x.val for x in self.ff.data])
writerows = [[],[],[],[]]
for data_type in data_types:
writerows[0].extend([x.lbl for x in r_dict[data_type]])
writerows[1].extend([x.wht for x in r_dict[data_type]])
writerows[2].extend([x.val for x in r_dict[data_type]])
writerows[3].extend([x.val for x in c_dict[data_type]])
for row in writerows:
csv_writer.writerow(row)
logger.log(20, '~~ DIFFERENTIATING PARAMETERS ~~'.rjust(79, '~'))
# Save many FFs, each with their own parameter sets.
ffs = opt.differentiate_ff(self.ff)
logger.log(20, '~~ SCORING DIFFERENTIATED PARAMETERS ~~'.rjust(
79, '~'))
for ff in ffs:
ff.export_ff(lines=self.ff.lines)
logger.log(20, ' -- Calculating {}.'.format(ff))
data = calculate.main(self.args_ff)
# Deprecated
#ff.score = compare.compare_data(ref_data, data)
c_data = compare.data_by_type(data)
r_dict, c_data = compare.trim_data(r_dict,c_data)
ff.score = compare.compare_data(r_dict, c_data)
opt.pretty_ff_results(ff)
# Write the data rather than storing it in memory. For large
# parameter sets, this could consume GBs of memory otherwise!
#csv_writer.writerow([x.val for x in data])
row = []
for data_type in data_types:
row.extend([x.val for x in c_data[data_type]])
csv_writer.writerow(row)
f.close()
# Make sure we have derivative information. Used for NR.
#
# The derivatives are useful for checking up on the progress of the
# optimization and for deciding which parameters to use in a
# subsequent simplex optimization.
#
# Still need a way to do this with the resatrt file.
opt.param_derivs(self.ff, ffs)
# Calculate the Jacobian, residual vector, matrix A and vector b.
# These aren't needed if you're only doing Newton-Raphson.
if self.do_lstsq or self.do_lagrange or self.do_levenberg or \
self.do_svd:
logger.log(20, '~~ JACOBIAN AND RESIDUAL VECTOR ~~'.rjust(79, '~'))
# Setup the residual vector.
# Deprecated - TR
#num_d = len(ref_data)
num_d = 0
for datatype in r_dict:
num_d += len(r_dict[datatype])
resid = np.empty((num_d, 1), dtype=float)
# Deprecated - TR
#for i in xrange(0, num_d):
# resid[i, 0] = ref_data[i].wht * \
# (ref_data[i].val - self.ff.data[i].val)
count = 0
for data_type in data_types:
for r,c in zip(r_dict[data_type],c_dict[data_type]):
resid[count, 0] = r.wht * (r.val - c.val)
count += 1
# logger.log(5, 'RESIDUAL VECTOR:\n{}'.format(resid))
logger.log(
20, ' -- Formed {} residual vector.'.format(resid.shape))
# Setup the Jacobian.
num_p = len(self.ff.params)
# Maybe should be a part of the Jacobian function.
jacob = np.empty((num_d, num_p), dtype=float)
jacob = return_jacobian(jacob, os.path.join(self.direc, par_file))
# logger.log(5, 'JACOBIAN:\n{}'.format(jacob))
logger.log(20, ' -- Formed {} Jacobian.'.format(jacob.shape))
ma = jacob.T.dot(jacob)
vb = jacob.T.dot(resid)
# We need these for most optimization methods.
logger.log(5, ' MATRIX A AND VECTOR B '.center(79, '-'))
# logger.log(5, 'A:\n{}'.format(ma))
# logger.log(5, 'b:\n{}'.format(vb))
# Start coming up with new parameter sets.
if self.do_newton and not restart:
logger.log(20, '~~ NEWTON-RAPHSON ~~'.rjust(79, '~'))
# Moved the derivative section outside of here.
changes = do_newton(self.ff.params,
radii=self.newton_radii,
cutoffs=self.newton_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
if self.do_lstsq:
logger.log(20, '~~ LEAST SQUARES ~~'.rjust(79, '~'))
changes = do_lstsq(ma, vb,
radii=self.lstsq_radii,
cutoffs=self.lstsq_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
if self.do_lagrange:
logger.log(20, '~~ LAGRANGE ~~'.rjust(79, '~'))
for factor in sorted(self.lagrange_factors):
changes = do_lagrange(ma, vb, factor,
radii=self.lagrange_radii,
cutoffs=self.lagrange_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
if self.do_levenberg:
logger.log(20, '~~ LEVENBERG ~~'.rjust(79, '~'))
for factor in sorted(self.levenberg_factors):
changes = do_levenberg(ma, vb, factor,
radii=self.levenberg_radii,
cutoffs=self.levenberg_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
if self.do_svd:
logger.log(20, '~~ SINGULAR VALUE DECOMPOSITION ~~'.rjust(79, '~'))
# J = U . s . VT
mu, vs, mvt = return_svd(jacob)
logger.log(1, '>>> mu.shape: {}'.format(mu.shape))
logger.log(1, '>>> vs.shape: {}'.format(vs.shape))
logger.log(1, '>>> mvt.shape: {}'.format(mvt.shape))
logger.log(1, '>>> vb.shape: {}'.format(vb.shape))
if self.svd_factors:
changes = do_svd_w_thresholds(mu, vs, mvt, resid, self.svd_factors,
radii=self.svd_radii,
cutoffs=self.svd_cutoffs)
else:
changes = do_svd_wo_thresholds(mu, vs, mvt, resid,
radii=self.svd_radii,
cutoffs=self.svd_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
# Report how many trial FFs were generated.
logger.log(20, ' -- Generated {} trial force field(s).'.format(
len(self.new_ffs)))
# If there are any trials, test them.
if self.new_ffs:
logger.log(20, '~~ EVALUATING TRIAL FF(S) ~~'.rjust(79, '~'))
for ff in self.new_ffs:
data = opt.cal_ff(ff, self.args_ff, parent_ff=self.ff)
# Shouldn't need to zero anymore.
# Deprecated
#ff.score = compare.compare_data(ref_data, data)
c_data = compare.data_by_type(data)
r_dict, c_data = compare.trim_data(r_dict,c_data)
ff.score = compare.compare_data(r_dict, c_data)
opt.pretty_ff_results(ff)
self.new_ffs = sorted(
self.new_ffs, key=lambda x: x.score)
# Check for improvement.
if self.new_ffs[0].score < self.ff.score:
ff = self.new_ffs[0]
logger.log(
20, '~~ GRADIENT FINISHED WITH IMPROVEMENTS ~~'.rjust(
79, '~'))
opt.pretty_ff_results(self.ff, level=20)
opt.pretty_ff_results(ff, level=20)
# Copy parameter derivatives from original FF to save time in
# case we move onto simplex immediately after this.
copy_derivs(self.ff, ff)
else:
ff = self.ff
else:
ff = self.ff
return ff
def run(self, ref_data=None):
# We need reference data if you didn't provide it.
if ref_data is None:
ref_data = opt.return_ref_data(self.args_ref)
# We need the initial FF data.
if self.ff.data is None:
logger.log(20, '~~ GATHERING INITIAL FF DATA ~~'.rjust(79, '~'))
# Check whether this is efficient with the ff_lines.
self.ff.export_ff()
self.ff.data = calculate.main(self.args_ff)
# We could do this, but the zeroing of energies has already been
# done.
# self.ff.score = compare.compare_data(ref_data, self.ff.data)
# So instead we do this.
compare.correlate_energies(ref_data, self.ff.data)
if self.ff.score is None:
# Already zeroed reference and correlated the energies.
self.ff.score = compare.calculate_score(ref_data, self.ff.data)
logger.log(20, 'INITIAL FF SCORE: {}'.format(self.ff.fscore))
logger.log(20, '~~ GRADIENT OPTIMIZATION ~~'.rjust(79, '~'))
# We need a file to hold the differentiated parameter data.
par_files = glob.glob(os.path.join(self.direc, 'par_diff_???.txt'))
if par_files:
par_files.sort()
most_recent_par_file = par_files[-1]
most_recent_par_file = most_recent_par_file.split('/')[-1]
most_recent_num = most_recent_par_file[9:12]
num = int(most_recent_num) + 1
par_file = 'par_diff_{:03d}.txt'.format(num)
else:
par_file = 'par_diff_001.txt'
f = open(os.path.join(self.direc, par_file), 'w')
csv_writer = csv.writer(f)
# Row 1 - Labels
# Row 2 - Weights
# Row 3 - Reference data values
# Row 4 - Initial FF data values
csv_writer.writerow([x.lbl for x in ref_data])
csv_writer.writerow([x.wht for x in ref_data])
csv_writer.writerow([x.val for x in ref_data])
csv_writer.writerow([x.val for x in self.ff.data])
logger.log(20, '~~ DIFFERENTIATING PARAMETERS ~~'.rjust(79, '~'))
# Setup the residual vector.
# Perhaps move this closer to the Jacobian section.
num_d = len(ref_data)
resid = np.empty((num_d, 1), dtype=float)
for i in xrange(0, num_d):
resid[i, 0] = ref_data[i].wht * (ref_data[i].val - self.ff.data[i].val)
logger.log(5, 'RESIDUAL VECTOR:\n{}'.format(resid))
logger.log(20, ' -- Formed {} residual vector.'.format(resid.shape))
# Save many FFs, each with their own parameter sets.
ffs = opt.differentiate_ff(self.ff)
logger.log(20, '~~ SCORING DIFFERENTIATED PARAMETERS ~~'.rjust(79, '~'))
for ff in ffs:
ff.export_ff(lines=self.ff.lines)
logger.log(20, ' -- Calculating {}.'.format(ff))
data = calculate.main(self.args_ff)
compare.correlate_energies(ref_data, data)
ff.score = compare.calculate_score(ref_data, data)
opt.pretty_ff_results(ff)
# Write the data rather than storing it in memory. For large parameter
# sets, this could consume GBs of memory otherwise!
csv_writer.writerow([x.val for x in data])
f.close()
# Calculate the Jacobian, residual vector, matrix A and vector b.
# These aren't needed if you're only doing Newton-Raphson.
if self.do_lstsq or self.do_lagrange or self.do_levenberg or \
self.do_svd:
logger.log(20, '~~ JACOBIAN AND RESIDUAL VECTOR ~~'.rjust(79, '~'))
# Setup the Jacobian.
num_p = len(self.ff.params)
# Maybe should be a part of the Jacobian function.
jacob = np.empty((num_d, num_p), dtype=float)
jacob = return_jacobian(jacob, os.path.join(self.direc, par_file))
ma = jacob.T.dot(jacob)
vb = jacob.T.dot(resid)
# We need these for most optimization methods.
logger.log(5, ' MATRIX A AND VECTOR B '.center(79, '-'))
logger.log(5, 'A:\n{}'.format(ma))
logger.log(5, 'b:\n{}'.format(vb))
# Start coming up with new parameter sets.
if self.do_newton:
logger.log(20, '~~ NEWTON-RAPHSON ~~'.rjust(79, '~'))
# Make sure we have derivative information.
if self.ff.params[0].d1 is None:
opt.param_derivs(self.ff, ffs)
changes = do_newton(self.ff.params,
radii=self.newton_radii,
cutoffs=self.newton_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
if self.do_lstsq:
logger.log(20, '~~ LEAST SQUARES ~~'.rjust(79, '~'))
changes = do_lstsq(ma, vb,
radii=self.lstsq_radii,
cutoffs=self.lstsq_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
if self.do_lagrange:
logger.log(20, '~~ LAGRANGE ~~'.rjust(79, '~'))
for factor in sorted(self.lagrange_factors):
changes = do_lagrange(ma, vb, factor,
radii=self.lagrange_radii,
cutoffs=self.lagrange_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
if self.do_levenberg:
logger.log(20, '~~ LEVENBERG ~~'.rjust(79, '~'))
for factor in sorted(self.levenberg_factors):
changes = do_levenberg(ma, vb, factor,
radii=self.levenberg_radii,
cutoffs=self.levenberg_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
if self.do_svd:
logger.log(20, '~~ SINGULAR VALUE DECOMPOSITION ~~'.rjust(79, '~'))
mu, vs, mv = return_svd(ma)
if self.svd_factors:
changes = do_svd_w_thresholds(mu, vs, mv, vb, self.svd_factors,
radii=self.svd_radii,
cutoffs=self.svd_cutoffs)
else:
changes = do_svd_wo_thresholds(mu, vs, mv, vb,
radii=self.svd_radii,
cutoffs=self.svd_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
# Report how many trial FFs were generated.
logger.log(20, ' -- Generated {} trial force field(s).'.format(
len(self.new_ffs)))
# If there are any trials, test them.
if self.new_ffs:
logger.log(20, '~~ EVALUATING TRIAL FF(S) ~~'.rjust(79, '~'))
for ff in self.new_ffs:
data = opt.cal_ff(ff, self.args_ff, parent_ff=self.ff)
ff.score = compare.compare_data(ref_data, data, zero=False)
opt.pretty_ff_results(ff)
self.new_ffs = sorted(
self.new_ffs, key=lambda x: x.score)
# Check for improvement.
if self.new_ffs[0].score < self.ff.score:
ff = self.new_ffs[0]
logger.log(
20, '~~ GRADIENT FINISHED WITH IMPROVEMENTS ~~'.rjust(
79, '~'))
opt.pretty_ff_results(self.ff, level=20)
opt.pretty_ff_results(ff, level=20)
else:
ff = self.ff
else:
ff = self.ff
return ff
def run(self, r_data=None):
"""
Once all attributes are setup as you so desire, run this method to
optimize the parameters.
Returns
-------
`datatypes.FF` (or subclass)
Contains the best parameters.
"""
if r_data is None:
r_data = opt.return_ref_data(self.args_ref)
logger.log(20, '~~ SIMPLEX OPTIMIZATION ~~'.rjust(79, '~'))
# Here we don't actually need the database connection/force field data.
# We only need the score.
if self.ff.score is None:
logger.log(20, '~~ CALCULATING INITIAL FF SCORE ~~'.rjust(79, '~'))
self.ff.export_ff()
# I could store this object to prevent on self.ff to prevent garbage
# collection. Would be nice if simplex was followed by gradient,
# which needs that information, and if simplex yielded no
# improvements. At most points in the optimization, this is probably
# too infrequent for it to be worth the memory, but it might be nice
# once the parameters are close to convergence.
data = calculate.main(self.args_ff)
self.ff.score = compare.compare_data(r_data, data)
logger.log(20, 'INITIAL FF SCORE: {}'.format(self.ff.score))
else:
logger.log(15,
' -- Reused existing score and data for initial FF.')
logger.log(15, 'INIT FF SCORE: {}'.format(self.ff.score))
if self.max_params and len(self.ff.params) > self.max_params:
if self.ff.params[0].d1:
logger.log(15, ' -- Reusing existing parameter derivatives.')
# Don't score so this really doesn't take much time.
ffs = opt.differentiate_ff(self.ff, central=False)
else:
logger.log(15, ' -- Calculating new parameter derivatives.')
ffs = opt.differentiate_ff(self.ff, central=True)
# We have to score to get the derivatives.
for ff in ffs:
ff.export_ff(lines=self.ff_lines)
logger.log(20, ' -- Calculating {}.'.format(ff))
data = calculate.main(self.args_ff)
ff.score = compare.compare_data(r_data, data)
opt.pretty_ff_results(ff)
opt.param_derivs(self.ff, ffs)
# Only keep the forward differentiated FFs.
ffs = opt.extract_forward(ffs)
params = select_simp_params_on_derivs(self.ff.params,
max_params=self.max_params)
self.new_ffs = opt.extract_ff_by_params(ffs, params)
# Reduce number of parameters.
# Will need an option that's not MM3* specific.
ff_rows = [x.mm3_row for x in params]
ff_cols = [x.mm3_col for x in params]
for ff in self.new_ffs:
new_params = []
for param in ff.params:
if param.mm3_row in ff_rows and param.mm3_col in ff_cols:
new_params.append(param)
ff.params = new_params
else:
self.new_ffs = opt.differentiate_ff(self.ff, central=False)
# Double check and make sure they're all scored.
for ff in self.new_ffs:
if ff.score is None:
ff.export_ff(lines=self.ff_lines)
logger.log(20, ' -- Calculating {}.'.format(ff))
data = calculate.main(self.args_ff)
ff.score = compare.compare_data(r_data, data)
opt.pretty_ff_results(ff)
ff_copy = copy.deepcopy(self.ff)
new_params = []
for param in ff.params:
if param.mm3_row in ff_rows and param.mm3_col in ff_cols:
new_params.append(param)
ff_copy.params = new_params
self.new_ffs = sorted(self.new_ffs + [ff_copy], key=lambda x: x.score)
wrapper = textwrap.TextWrapper(width=79)
logger.log(20, 'ORDERED FF SCORES:')
logger.log(
20,
wrapper.fill('{}'.format(' '.join('{:15.4f}'.format(x.score)
for x in self.new_ffs))))
# Shows all FFs parameters.
opt.pretty_ff_params(self.new_ffs)
# Start the simplex cycles.
current_cycle = 0
cycles_wo_change = 0
while current_cycle < self.max_cycles \
and cycles_wo_change < self.max_cycles_wo_change:
current_cycle += 1
last_best = self.new_ffs[0].score
best_ff = self.new_ffs[0]
logger.log(
20, '~~ START SIMPLEX CYCLE {} ~~'.format(current_cycle).rjust(
79, '~'))
inv_ff = self.ff.__class__()
if self.do_weighted_reflection:
inv_ff.method = 'WEIGHTED INVERSION'
else:
inv_ff.method = 'INVERSION'
inv_ff.params = copy.deepcopy(best_ff.params)
ref_ff = self.ff.__class__()
ref_ff.method = 'REFLECTION'
ref_ff.params = copy.deepcopy(best_ff.params)
for i in xrange(0, len(best_ff.params)):
if self.do_weighted_reflection:
try:
inv_val = (sum([
x.params[i].value *
(x.score - self.new_ffs[-1].score)
for x in self.new_ffs[:-1]
]) / sum([
x.score - self.new_ffs[-1].score
for x in self.new_ffs[:-1]
]))
except ZeroDivisionError:
logger.warning(
'Attempted to divide by zero while calculating the '
'weighted simplex inversion point. All penalty '
'function scores for the trial force fields are '
'numerically equivalent.')
# Breaking should just exit the while loop. Should still
# give you the best force field determined thus far.
break
else:
inv_val = (
sum([x.params[i].value for x in self.new_ffs[:-1]]) /
len(self.new_ffs[:-1]))
inv_ff.params[i].value = inv_val
ref_ff.params[i].value = (2 * inv_val -
self.new_ffs[-1].params[i].value)
# Calculate score for inverted parameters.
self.ff.export_ff(self.ff.path, params=inv_ff.params)
data = calculate.main(self.args_ff)
inv_ff.score = compare.compare_data(r_data, data)
opt.pretty_ff_results(inv_ff)
# Calculate score for reflected parameters.
self.ff.export_ff(self.ff.path, params=ref_ff.params)
data = calculate.main(self.args_ff)
ref_ff.score = compare.compare_data(r_data, data)
opt.pretty_ff_results(ref_ff)
if ref_ff.score < self.new_ffs[0].score:
logger.log(20, '~~ ATTEMPTING EXPANSION ~~'.rjust(79, '~'))
exp_ff = self.ff.__class__()
exp_ff.method = 'EXPANSION'
exp_ff.params = copy.deepcopy(best_ff.params)
for i in xrange(0, len(self.new_ffs[0].params)):
exp_ff.params[i].value = (
3 * inv_ff.params[i].value -
2 * self.new_ffs[-1].params[i].value)
self.ff.export_ff(self.ff.path, exp_ff.params)
data = calculate.main(self.args_ff)
exp_ff.score = compare.compare_data(r_data, data)
opt.pretty_ff_results(exp_ff)
if exp_ff.score < ref_ff.score:
self.new_ffs[-1] = exp_ff
logger.log(
20, ' -- Expansion succeeded. Keeping expanded '
'parameters.')
else:
self.new_ffs[-1] = ref_ff
logger.log(
20,
' -- Expansion failed. Keeping reflected parameters.')
elif ref_ff.score < self.new_ffs[-2].score:
logger.log(20, ' -- Keeping reflected parameters.')
self.new_ffs[-1] = ref_ff
else:
logger.log(20, '~~ ATTEMPTING CONTRACTION ~~'.rjust(79, '~'))
con_ff = self.ff.__class__()
con_ff.method = 'CONTRACTION'
con_ff.params = copy.deepcopy(best_ff.params)
for i in xrange(0, len(best_ff.params)):
if ref_ff.score > self.new_ffs[-1].score:
con_val = ((inv_ff.params[i].value +
self.new_ffs[-1].params[i].value) / 2)
else:
con_val = ((3 * inv_ff.params[i].value -
self.new_ffs[-1].params[i].value) / 2)
con_ff.params[i].value = con_val
self.ff.export_ff(self.ff.path, params=con_ff.params)
data = calculate.main(self.args_ff)
con_ff.score = compare.compare_data(r_data, data)
opt.pretty_ff_results(con_ff)
if con_ff.score < self.new_ffs[-2].score:
self.new_ffs[-1] = con_ff
elif self.do_massive_contraction:
logger.log(
20, '~~ DOING MASSIVE CONTRACTION ~~'.rjust(79, '~'))
for ff_num, ff in enumerate(self.new_ffs[1:]):
for i in xrange(0, len(best_ff.params)):
ff.params[i].value = (
(ff.params[i].value +
self.new_ffs[0].params[i].value) / 2)
self.ff.export_ff(self.ff.path, params=ff.params)
data = calculate.main(self.args_ff)
ff.score = compare.compare_data(r_data, data)
ff.method += ' MC'
opt.pretty_ff_results(ff)
else:
logger.log(20, ' -- Contraction failed.')
self.new_ffs = sorted(self.new_ffs, key=lambda x: x.score)
if self.new_ffs[0].score < last_best:
cycles_wo_change = 0
else:
cycles_wo_change += 1
logger.log(
20,
' -- {} cycles without change.'.format(cycles_wo_change))
best_ff = self.new_ffs[0]
logger.log(20, 'BEST:')
opt.pretty_ff_results(self.new_ffs[0], level=20)
logger.log(
20, '~~ END SIMPLEX CYCLE {} ~~'.format(current_cycle).rjust(
79, '~'))
if best_ff.score < self.ff.score:
logger.log(
20, '~~ SIMPLEX FINISHED WITH IMPROVEMENTS ~~'.rjust(79, '~'))
best_ff = restore_simp_ff(best_ff, self.ff)
else:
logger.log(
20,
'~~ SIMPLEX FINISHED WITHOUT IMPROVEMENTS ~~'.rjust(79, '~'))
opt.pretty_ff_results(self.ff, level=20)
opt.pretty_ff_results(best_ff, level=20)
logger.log(20, ' -- Writing best force field from simplex.')
best_ff.export_ff(best_ff.path)
return best_ff
def run(self, r_data=None):
"""
Once all attributes are setup as you so desire, run this method to
optimize the parameters.
Returns
-------
`datatypes.FF` (or subclass)
Contains the best parameters.
"""
if r_data is None:
r_data = opt.return_ref_data(self.args_ref)
if self.ff.score is None:
logger.log(20, '~~ CALCULATING INITIAL FF SCORE ~~'.rjust(79, '~'))
self.ff.export_ff()
# Could store data on self.ff.data if we wanted. Not necessary for
# simplex. If simplex yielded no improvements, it would return this
# FF, and then we might want the data such taht we don't have to
# recalculate it in gradient. Let's hope simplex generally yields
# improvements.
data = calculate.main(self.args_ff)
#deprecated
#self.ff.score = compare.compare_data(r_data, data)
r_dict = compare.data_by_type(r_data)
c_dict = compare.data_by_type(data)
r_dict, c_dict = compare.trim_data(r_dict, c_dict)
self.ff.score = compare.compare_data(r_dict, c_dict)
else:
logger.log(20,
' -- Reused existing score and data for initial FF.')
logger.log(20, '~~ SIMPLEX OPTIMIZATION ~~'.rjust(79, '~'))
logger.log(20, 'INIT FF SCORE: {}'.format(self.ff.score))
opt.pretty_ff_results(self.ff, level=20)
# Here's what we do if there are too many parameters.
if self.max_params and len(self.ff.params) > self.max_params:
logger.log(20, ' -- More parameters than the maximum allowed.')
logger.log(5, 'CURRENT PARAMS: {}'.format(len(self.ff.params)))
logger.log(5, 'MAX PARAMS: {}'.format(self.max_params))
# Here we select the parameters that have the lowest 2nd
# derivatives.
# Could fail when simplex finds improvements but restores other
# parameters.
# if self.ff.params[0].d1:
if None in [x.d1 for x in self.ff.params]:
logger.log(15, ' -- Calculating new parameter derivatives.')
# Do central differentiation so we can calculate derivatives.
# Another option would be to write code to determine
# derivatives only from forward differentiation.
ffs = opt.differentiate_ff(self.ff, central=True)
# We have to score to get the derivatives.
for ff in ffs:
ff.export_ff(path=self.ff.path, lines=self.ff_lines)
logger.log(20, ' -- Calculating {}.'.format(ff))
data = calculate.main(self.args_ff)
#deprecated
#ff.score = compare.compare_data(r_data, data)
r_dict = compare.data_by_type(r_data)
c_dict = compare.data_by_type(data)
r_dict, c_dict = compare.trim_data(r_dict, c_dict)
ff.score = compare.compare_data(r_dict, c_dict)
opt.pretty_ff_results(ff)
# Add the derivatives to your original FF.
opt.param_derivs(self.ff, ffs)
# Only keep the forward differentiated FFs.
ffs = opt.extract_forward(ffs)
logger.log(
5, ' -- Keeping {} forward differentiated '
'FFs.'.format(len(ffs)))
else:
logger.log(15, ' -- Reusing existing parameter derivatives.')
# Differentiate all parameters forward. Yes, I know this is
# counter-intuitive because we are going to only use subset of
# the forward differentiated FFs. However, this is very
# computationally inexpensive because we're not scoring them
# now. We will remove the forward differentiated FFs we don't
# want before scoring.
ffs = opt.differentiate_ff(self.ff, central=False)
# This sorts the parameters based upon their 2nd derivative.
# It keeps the ones with lowest 2nd derivatives.
# SCHEDULED FOR CHANGES. NOT A GOOD SORTING CRITERION.
params = select_simp_params_on_derivs(self.ff.params,
max_params=self.max_params)
# From the entire list of forward differentiated FFs, pick
# out the ones that have the lowest 2nd derivatives.
self.new_ffs = opt.extract_ff_by_params(ffs, params)
logger.log(1,
'>>> len(self.new_ffs): {}'.format(len(self.new_ffs)))
# Reduce number of parameters.
# Will need an option that's not MM3* specific in the future.
ff_rows = [x.mm3_row for x in params]
ff_cols = [x.mm3_col for x in params]
for ff in self.new_ffs:
new_params = []
for param in ff.params:
if param.mm3_row in ff_rows and param.mm3_col in ff_cols:
new_params.append(param)
ff.params = new_params
# Make a copy of your original FF that has less parameters.
ff_copy = copy.deepcopy(self.ff)
new_params = []
for param in ff.params:
if param.mm3_row in ff_rows and param.mm3_col in ff_cols:
new_params.append(param)
ff_copy.params = new_params
else:
# In this case it's simple. Just forward differentiate each
# parameter.
self.new_ffs = opt.differentiate_ff(self.ff, central=False)
logger.log(1,
'>>> len(self.new_ffs): {}'.format(len(self.new_ffs)))
# Still make that FF copy.
ff_copy = copy.deepcopy(self.ff)
# Double check and make sure they're all scored.
for ff in self.new_ffs:
if ff.score is None:
ff.export_ff(path=self.ff.path, lines=self.ff_lines)
logger.log(20, ' -- Calculating {}.'.format(ff))
data = calculate.main(self.args_ff)
#deprecated
#ff.score = compare.compare_data(r_data, data)
r_dict = compare.data_by_type(r_data)
c_dict = compare.data_by_type(data)
r_dict, c_dict = compare.trim_data(r_dict, c_dict)
ff.score = compare.compare_data(r_dict, c_dict)
opt.pretty_ff_results(ff)
# Add your copy of the orignal to FF to the forward differentiated FFs.
self.new_ffs = sorted(self.new_ffs + [ff_copy], key=lambda x: x.score)
# Allow 3 cycles w/o change for each parameter present. Remember that
# the initial FF was added here, hence the minus one.
self._max_cycles_wo_change = 3 * (len(self.new_ffs) - 1)
wrapper = textwrap.TextWrapper(width=79)
# Shows all FFs parameters.
opt.pretty_ff_params(self.new_ffs)
# Start the simplex cycles.
current_cycle = 0
cycles_wo_change = 0
while current_cycle < self.max_cycles \
and cycles_wo_change < self._max_cycles_wo_change:
current_cycle += 1
# Save the last best in case some accidental sort goes on.
# Plus it makes reading the code a litle easier.
last_best_ff = copy.deepcopy(self.new_ffs[0])
logger.log(
20, '~~ START SIMPLEX CYCLE {} ~~'.format(current_cycle).rjust(
79, '~'))
logger.log(20, 'ORDERED FF SCORES:')
logger.log(
20,
wrapper.fill('{}'.format(' '.join('{:15.4f}'.format(x.score)
for x in self.new_ffs))))
inv_ff = self.ff.__class__()
if self.do_weighted_reflection:
inv_ff.method = 'WEIGHTED INVERSION'
else:
inv_ff.method = 'INVERSION'
inv_ff.params = copy.deepcopy(last_best_ff.params)
ref_ff = self.ff.__class__()
ref_ff.method = 'REFLECTION'
ref_ff.params = copy.deepcopy(last_best_ff.params)
# Need score difference sum for weighted inversion.
# Calculate this value before going into loop.
if self.do_weighted_reflection:
# If zero, should break.
score_diff_sum = sum([
x.score - self.new_ffs[-1].score for x in self.new_ffs[:-1]
])
if score_diff_sum == 0.:
logger.warning('No difference between force field scores. '
'Exiting simplex.')
# We want to raise opt.OptError such that
# opt.catch_run_errors will write the best FF obtained thus
# far.
raise opt.OptError(
'No difference between force field scores. '
'Exiting simplex.')
for i in range(0, len(last_best_ff.params)):
if self.do_weighted_reflection:
inv_val = (sum([
x.params[i].value * (x.score - self.new_ffs[-1].score)
for x in self.new_ffs[:-1]
]) / score_diff_sum)
else:
inv_val = (
sum([x.params[i].value for x in self.new_ffs[:-1]]) /
len(self.new_ffs[:-1]))
inv_ff.params[i].value = inv_val
ref_ff.params[i].value = (2 * inv_val -
self.new_ffs[-1].params[i].value)
# The inversion point does not need to be scored.
# Calculate score for reflected parameters.
ref_ff.export_ff(path=self.ff.path, lines=self.ff.lines)
data = calculate.main(self.args_ff)
#deprecated
#ref_ff.score = compare.compare_data(r_data, data)
r_dict = compare.data_by_type(r_data)
c_dict = compare.data_by_type(data)
r_dict, c_dict = compare.trim_data(r_dict, c_dict)
ref_ff.score = compare.compare_data(r_dict, c_dict)
opt.pretty_ff_results(ref_ff)
if ref_ff.score < last_best_ff.score:
logger.log(20, '~~ ATTEMPTING EXPANSION ~~'.rjust(79, '~'))
exp_ff = self.ff.__class__()
exp_ff.method = 'EXPANSION'
exp_ff.params = copy.deepcopy(last_best_ff.params)
for i in range(0, len(last_best_ff.params)):
exp_ff.params[i].value = (
3 * inv_ff.params[i].value -
2 * self.new_ffs[-1].params[i].value)
exp_ff.export_ff(path=self.ff.path, lines=self.ff.lines)
data = calculate.main(self.args_ff)
#deprecated
#exp_ff.score = compare.compare_data(r_data, data)
r_dict = compare.data_by_type(r_data)
c_dict = compare.data_by_type(data)
r_dict, c_dict = compare.trim_data(r_dict, c_dict)
exp_ff.score = compare.compare_data(r_dict, c_dict)
opt.pretty_ff_results(exp_ff)
if exp_ff.score < ref_ff.score:
self.new_ffs[-1] = exp_ff
logger.log(
20, ' -- Expansion succeeded. Keeping expanded '
'parameters.')
else:
self.new_ffs[-1] = ref_ff
logger.log(
20,
' -- Expansion failed. Keeping reflected parameters.')
elif ref_ff.score < self.new_ffs[-2].score:
logger.log(20, ' -- Keeping reflected parameters.')
self.new_ffs[-1] = ref_ff
else:
logger.log(20, '~~ ATTEMPTING CONTRACTION ~~'.rjust(79, '~'))
con_ff = self.ff.__class__()
con_ff.method = 'CONTRACTION'
con_ff.params = copy.deepcopy(last_best_ff.params)
for i in range(0, len(last_best_ff.params)):
if ref_ff.score > self.new_ffs[-1].score:
con_val = ((inv_ff.params[i].value +
self.new_ffs[-1].params[i].value) / 2)
else:
con_val = ((3 * inv_ff.params[i].value -
self.new_ffs[-1].params[i].value) / 2)
con_ff.params[i].value = con_val
self.ff.export_ff(params=con_ff.params)
data = calculate.main(self.args_ff)
#deprecated
#con_ff.score = compare.compare_data(r_data, data)
r_dict = compare.data_by_type(r_data)
c_dict = compare.data_by_type(data)
r_dict, c_dict = compare.trim_data(r_dict, c_dict)
con_ff.score = compare.compare_data(r_dict, c_dict)
opt.pretty_ff_results(con_ff)
# This change was made to reflect the 1998 Q2MM publication.
# if con_ff.score < self.new_ffs[-1].score:
if con_ff.score < self.new_ffs[-2].score:
logger.log(20, ' -- Contraction succeeded.')
self.new_ffs[-1] = con_ff
elif self.do_massive_contraction:
logger.log(
20, '~~ DOING MASSIVE CONTRACTION ~~'.rjust(79, '~'))
for ff_num, ff in enumerate(self.new_ffs[1:]):
for i in range(0, len(last_best_ff.params)):
ff.params[i].value = (
(ff.params[i].value +
self.new_ffs[0].params[i].value) / 2)
self.ff.export_ff(params=ff.params)
data = calculate.main(self.args_ff)
#deprecated
#ff.score = compare.compare_data(r_data, data)
r_dict = compare.data_by_type(r_data)
c_dict = compare.data_by_type(data)
r_dict, c_dict = compare.trim_data(r_dict, c_dict)
ff.score = compare.compare_data(r_dict, c_dict)
ff.method += ' MC'
opt.pretty_ff_results(ff)
else:
logger.log(
20, ' -- Contraction failed. Keeping parmaeters '
'anyway.')
self.new_ffs[-1] = con_ff
self.new_ffs = sorted(self.new_ffs, key=lambda x: x.score)
# Keep track of the number of cycles without change. If there's
# improvement, reset the counter.
if self.new_ffs[0].score < last_best_ff.score:
cycles_wo_change = 0
else:
cycles_wo_change += 1
logger.log(
20, ' -- {} cycles without improvement out of {} '
'allowed.'.format(cycles_wo_change,
self._max_cycles_wo_change))
logger.log(20, 'BEST:')
opt.pretty_ff_results(self.new_ffs[0], level=20)
logger.log(
20, '~~ END SIMPLEX CYCLE {} ~~'.format(current_cycle).rjust(
79, '~'))
# This sort is likely unnecessary because it should be done at the end
# of the last loop cycle, but I put it here just in case.
self.new_ffs = sorted(self.new_ffs, key=lambda x: x.score)
best_ff = self.new_ffs[0]
if best_ff.score < self.ff.score:
logger.log(
20, '~~ SIMPLEX FINISHED WITH IMPROVEMENTS ~~'.rjust(79, '~'))
best_ff = restore_simp_ff(best_ff, self.ff)
else:
logger.log(
20,
'~~ SIMPLEX FINISHED WITHOUT IMPROVEMENTS ~~'.rjust(79, '~'))
# This restores the inital parameters, so no need to use
# restore_simp_ff here.
best_ff = self.ff
opt.pretty_ff_results(self.ff, level=20)
opt.pretty_ff_results(best_ff, level=20)
logger.log(20, ' -- Writing best force field from simplex.')
best_ff.export_ff(best_ff.path)
return best_ff
def run(self, ref_data=None, restart=None):
# We need reference data if you didn't provide it.
if ref_data is None:
ref_data = opt.return_ref_data(self.args_ref)
# We need the initial FF data.
if self.ff.data is None:
logger.log(20, '~~ GATHERING INITIAL FF DATA ~~'.rjust(79, '~'))
# Check whether this is efficient with the ff_lines.
self.ff.export_ff()
self.ff.data = calculate.main(self.args_ff)
# We could do this, but the zeroing of energies has already been
# done.
# self.ff.score = compare.compare_data(ref_data, self.ff.data)
# So instead we do this.
compare.correlate_energies(ref_data, self.ff.data)
if self.ff.score is None:
# Already zeroed reference and correlated the energies.
self.ff.score = compare.calculate_score(ref_data, self.ff.data)
logger.log(20, 'INITIAL FF SCORE: {}'.format(self.ff.score))
logger.log(20, '~~ GRADIENT OPTIMIZATION ~~'.rjust(79, '~'))
# We need a file to hold the differentiated parameter data.
logger.log(20, '~~ CENTRAL DIFFERENTIATION ~~'.rjust(79, '~'))
if restart:
par_file = restart
logger.log(
20, ' -- Restarting gradient from central '
'differentiation file {}.'.format(par_file))
else:
par_files = glob.glob(os.path.join(self.direc, 'par_diff_???.txt'))
if par_files:
par_files.sort()
most_recent_par_file = par_files[-1]
most_recent_par_file = most_recent_par_file.split('/')[-1]
most_recent_num = most_recent_par_file[9:12]
num = int(most_recent_num) + 1
par_file = 'par_diff_{:03d}.txt'.format(num)
else:
par_file = 'par_diff_001.txt'
logger.log(
20, ' -- Generating central differentiation '
'file {}.'.format(par_file))
f = open(os.path.join(self.direc, par_file), 'w')
csv_writer = csv.writer(f)
# Row 1 - Labels
# Row 2 - Weights
# Row 3 - Reference data values
# Row 4 - Initial FF data values
csv_writer.writerow([x.lbl for x in ref_data])
csv_writer.writerow([x.wht for x in ref_data])
csv_writer.writerow([x.val for x in ref_data])
csv_writer.writerow([x.val for x in self.ff.data])
logger.log(20, '~~ DIFFERENTIATING PARAMETERS ~~'.rjust(79, '~'))
# Save many FFs, each with their own parameter sets.
ffs = opt.differentiate_ff(self.ff)
logger.log(
20, '~~ SCORING DIFFERENTIATED PARAMETERS ~~'.rjust(79, '~'))
for ff in ffs:
ff.export_ff(lines=self.ff.lines)
logger.log(20, ' -- Calculating {}.'.format(ff))
data = calculate.main(self.args_ff)
compare.correlate_energies(ref_data, data)
ff.score = compare.calculate_score(ref_data, data)
opt.pretty_ff_results(ff)
# Write the data rather than storing it in memory. For large parameter
# sets, this could consume GBs of memory otherwise!
csv_writer.writerow([x.val for x in data])
f.close()
# Calculate the Jacobian, residual vector, matrix A and vector b.
# These aren't needed if you're only doing Newton-Raphson.
if self.do_lstsq or self.do_lagrange or self.do_levenberg or \
self.do_svd:
logger.log(20, '~~ JACOBIAN AND RESIDUAL VECTOR ~~'.rjust(79, '~'))
# Setup the residual vector.
num_d = len(ref_data)
resid = np.empty((num_d, 1), dtype=float)
for i in xrange(0, num_d):
resid[i, 0] = ref_data[i].wht * (ref_data[i].val -
self.ff.data[i].val)
# logger.log(5, 'RESIDUAL VECTOR:\n{}'.format(resid))
logger.log(20,
' -- Formed {} residual vector.'.format(resid.shape))
# Setup the Jacobian.
num_p = len(self.ff.params)
# Maybe should be a part of the Jacobian function.
jacob = np.empty((num_d, num_p), dtype=float)
jacob = return_jacobian(jacob, os.path.join(self.direc, par_file))
# logger.log(5, 'JACOBIAN:\n{}'.format(jacob))
logger.log(20, ' -- Formed {} Jacobian.'.format(jacob.shape))
ma = jacob.T.dot(jacob)
vb = jacob.T.dot(resid)
# We need these for most optimization methods.
logger.log(5, ' MATRIX A AND VECTOR B '.center(79, '-'))
# logger.log(5, 'A:\n{}'.format(ma))
# logger.log(5, 'b:\n{}'.format(vb))
# Start coming up with new parameter sets.
if self.do_newton and not restart:
logger.log(20, '~~ NEWTON-RAPHSON ~~'.rjust(79, '~'))
# Make sure we have derivative information.
if self.ff.params[0].d1 is None:
opt.param_derivs(self.ff, ffs)
changes = do_newton(self.ff.params,
radii=self.newton_radii,
cutoffs=self.newton_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
if self.do_lstsq:
logger.log(20, '~~ LEAST SQUARES ~~'.rjust(79, '~'))
changes = do_lstsq(ma,
vb,
radii=self.lstsq_radii,
cutoffs=self.lstsq_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
if self.do_lagrange:
logger.log(20, '~~ LAGRANGE ~~'.rjust(79, '~'))
for factor in sorted(self.lagrange_factors):
changes = do_lagrange(ma,
vb,
factor,
radii=self.lagrange_radii,
cutoffs=self.lagrange_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
if self.do_levenberg:
logger.log(20, '~~ LEVENBERG ~~'.rjust(79, '~'))
for factor in sorted(self.levenberg_factors):
changes = do_levenberg(ma,
vb,
factor,
radii=self.levenberg_radii,
cutoffs=self.levenberg_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
if self.do_svd:
logger.log(20, '~~ SINGULAR VALUE DECOMPOSITION ~~'.rjust(79, '~'))
# J = U . s . VT
mu, vs, mvt = return_svd(jacob)
logger.log(1, '>>> mu.shape: {}'.format(mu.shape))
logger.log(1, '>>> vs.shape: {}'.format(vs.shape))
logger.log(1, '>>> mvt.shape: {}'.format(mvt.shape))
logger.log(1, '>>> vb.shape: {}'.format(vb.shape))
if self.svd_factors:
changes = do_svd_w_thresholds(mu,
vs,
mvt,
resid,
self.svd_factors,
radii=self.svd_radii,
cutoffs=self.svd_cutoffs)
else:
changes = do_svd_wo_thresholds(mu,
vs,
mvt,
resid,
radii=self.svd_radii,
cutoffs=self.svd_cutoffs)
cleanup(self.new_ffs, self.ff, changes)
# Report how many trial FFs were generated.
logger.log(
20, ' -- Generated {} trial force field(s).'.format(
len(self.new_ffs)))
# If there are any trials, test them.
if self.new_ffs:
logger.log(20, '~~ EVALUATING TRIAL FF(S) ~~'.rjust(79, '~'))
for ff in self.new_ffs:
data = opt.cal_ff(ff, self.args_ff, parent_ff=self.ff)
# Shouldn't need to zero anymore.
ff.score = compare.compare_data(ref_data, data)
opt.pretty_ff_results(ff)
self.new_ffs = sorted(self.new_ffs, key=lambda x: x.score)
# Check for improvement.
if self.new_ffs[0].score < self.ff.score:
ff = self.new_ffs[0]
logger.log(
20,
'~~ GRADIENT FINISHED WITH IMPROVEMENTS ~~'.rjust(79, '~'))
opt.pretty_ff_results(self.ff, level=20)
opt.pretty_ff_results(ff, level=20)
# Copy parameter derivatives from original FF to save time in
# case we move onto simplex immediately after this.
copy_derivs(self.ff, ff)
else:
ff = self.ff
else:
ff = self.ff
return ff