def run(self, summary_file_name, n_trajs_per_block=None):
# this will basically create the move_info_list for part of the
# summary_file, and then call super's RUN
summary = open(summary_file_name, 'r')
if self.summary_root_dir is None:
self.summary_root_dir = os.path.dirname(summary_file_name)
lines = [l for l in summary]
n_steps = len(lines)
if n_trajs_per_block is None:
n_trajs_per_block = n_steps
line_num = 0
while line_num < n_steps:
if self.report_progress is not None:
refresh_output("Working on MC step " + str(line_num) + "\n",
output_stream=self.report_progress)
end = min(line_num + n_trajs_per_block, n_steps)
block = lines[line_num:end]
moves = [self.parse_summary_line(l) for l in block]
super(OneWayTPSConverter, self).run(moves)
line_num += n_trajs_per_block
def run(self, summary_file_name, n_trajs_per_block=None):
# this will basically create the move_info_list for part of the
# summary_file, and then call super's RUN
summary = open(summary_file_name, 'r')
if self.summary_root_dir is None:
self.summary_root_dir = os.path.dirname(summary_file_name)
lines = [l for l in summary]
n_steps = len(lines)
if n_trajs_per_block is None:
n_trajs_per_block = n_steps
line_num = 0
while line_num < n_steps:
if self.report_progress is not None:
refresh_output("Working on MC step " + str(line_num) + "\n",
output_stream=self.report_progress)
end = min(line_num + n_trajs_per_block, n_steps)
block = lines[line_num:end]
moves = [self.parse_summary_line(l) for l in block]
super(OneWayTPSConverter, self).run(moves)
line_num += n_trajs_per_block
def generate_from_trajectories(
self,
ensembles,
trajectories,
preconditions=None,
strategies=None,
reuse_strategy='avoid',
engine=None):
"""
Create a SampleSet with as many initial samples as possible.
The goal of this is to give the initial SampleSet that would be
desired.
Parameters
----------
trajectories : list of :class:`.Trajectory` or :class:`.Trajectory`
the input trajectories to use
ensembles : list of :class:`Ensemble` or list of :class:`Ensemble`
the list of ensembles to be generated. If an element is itself a
list then one sample for one of the ensembles in that list if
generated
preconditions : list of str
a list of possible steps to modify the initial list of trajectories.
possible choices are
1. `sort-shortest` - sorting by shortest first,
2. `sort_median` - sorting by the middle one first and then in
move away from the median length
3. `sort-longest` - sorting by the longest first
4. `reverse` - reverse the order and
5. `mirror` which will add the reversed trajectories to the
list in the same order
Default is `None` which means to do nothing.
strategies : dict
a dict that specifies the options used when ensemble functions
are used to create a new sample.
reuse_strategy : str
if `avoid` then in a second attempt the used trajectories are
tried
engine : :class:`openpathsampling.engines.DyanmicsEngine`
the engine used for extending moves
Returns
-------
:class:`.SampleSet`
sampleset with samples for every initial ensemble for this
scheme that could be satisfied by the given trajectories
See Also
--------
list_initial_ensembles
"""
implemented_strategies = [
'get', # look for existing trajectories
'split', # look for existing sub-trajectories
'extend-complex', # try to extend long sub-trajectories
'extend-minimal' # try to extend short sub-trajectories
]
implemented_preconditions = [
'sort-shortest',
'sort-median',
'sort-longest',
'reverse',
'mirror'
]
if preconditions is None:
preconditions = ['mirror']
# create a list of trajectories
trajectories = paths.Trajectory._to_list_of_trajectories(trajectories)
for pre in preconditions:
if pre not in implemented_preconditions:
raise RuntimeError(
'%s is not a valid precondition strategy. Choose from %s.' %
(pre, implemented_preconditions)
)
if pre == 'sort-shortest':
trajectories = sorted(trajectories, key=len)
elif pre == 'sort-longest':
trajectories = sorted(trajectories, key=len)
elif pre == 'sort-median':
sorted_trajectories = sorted(trajectories, key=len)
trajectories = list([p for p2 in zip(
sorted_trajectories[len(sorted_trajectories) / 2:],
reversed(sorted_trajectories[:len(sorted_trajectories) / 2])
) for p in p2])
if len(sorted_trajectories) & 1:
trajectories.append(sorted_trajectories[-1])
elif pre == 'reverse':
trajectories = list(reversed(trajectories))
elif pre == 'mirror':
trajectories = trajectories + \
[traj.reversed for traj in trajectories]
# let's always try the short trajectories first
# print map(lambda x: hex(id(x)), trajectories)
# we will try forward/backward interleaved
used_trajectories = []
# if we start with an existing sample set look at what we got
# if we avoid we move the used ones to the back of the list
# if we remove we remove the used ones
for s in self:
traj = s.trajectory
if traj in trajectories:
used_trajectories.append(traj)
used_trajectories = sorted(used_trajectories, key=len)
# print map(lambda x: hex(id(x)), used_trajectories)
ensembles = [[x] if type(x) is not list else x for x in ensembles]
# 1. look in the existing sample_set
ensembles_to_fill, extra_ensembles = self.check_ensembles(ensembles)
# we reverse because we want to be able to remove elements
# from the list as we discover samples. This is easier to do
# when the list is traversed backwards since indices to not
# change, hence we reverse the list of ensemble and then traverse it
# in reversed order
ensembles_to_fill = \
list(reversed(ensembles_to_fill))
# 2. try strategies
if strategies is None:
# this is the default
strategies = [
'get',
'split'
]
for idx, strategy in enumerate(strategies):
if type(strategy) is str:
strategies[idx] = (strategy, dict())
if strategies[idx][0] not in implemented_strategies:
raise RuntimeError(
'Strategy `%s` is not known. Chose from %s.' % (
strategies[idx][0],
implemented_strategies
)
)
found_samples_str = ''
for pos, ens_list in enumerate(ensembles):
found_samples_str += '.' if ens_list in ensembles_to_fill else '+'
for str_idx, (strategy, options) in enumerate(strategies):
for idx, ens_list in reversed(list(enumerate(ensembles_to_fill))):
pos = ensembles.index(ens_list)
found_samples_str = \
found_samples_str[:pos] + \
'?' + found_samples_str[pos + 1:]
refresh_output((
'# trying strategy #%d `%s`: still missing %d samples\n'
'%s\n'
) % (
str_idx + 1,
strategy,
len(ensembles_to_fill),
found_samples_str
), ipynb_display_only=True, print_anyway=False)
if type(ens_list) is not list:
ens_list = [ens_list]
found = False
for ens in ens_list:
# create the list of options to be passed on
opts = {key: value for key, value in options.items()
if key not in ['exclude']}
# exclude contains the Ensemble classes to be ignored
if 'exclude' in options:
if isinstance(ens, options['exclude']):
continue
# fill only the first in ens_list that can be filled
sample = None
if strategy == 'get':
sample = ens.get_sample_from_trajectories(
trajectories=trajectories,
used_trajectories=used_trajectories,
reuse_strategy=reuse_strategy,
**opts
)
elif strategy == 'split':
sample = ens.split_sample_from_trajectories(
trajectories=trajectories,
used_trajectories=used_trajectories,
reuse_strategy=reuse_strategy,
**opts
)
elif strategy == 'extend-complex' and engine:
if hasattr(ens, 'extend_sample_from_trajectories'):
sample = ens.extend_sample_from_trajectories(
trajectories=trajectories,
engine=engine,
level='complex',
**opts
)
elif strategy == 'extend-minimal' and engine:
if hasattr(ens, 'extend_sample_from_trajectories'):
sample = ens.extend_sample_from_trajectories(
trajectories=trajectories,
engine=engine,
level='minimal',
**opts
)
elif strategy == 'extend-native' and engine:
if hasattr(ens, 'extend_sample_from_trajectories'):
sample = ens.extend_sample_from_trajectories(
trajectories=trajectories,
engine=engine,
level='native',
**opts
)
# now, if we've found a sample, add it and
# make sure we chose a proper replica ID
if sample is not None:
found = True
# another way would be to look for the smallest not
# taken id. This one is simpler
if len(self.replicas) > 0:
replica_idx = max(0, max(self.replicas) + 1)
else:
replica_idx = 0
sample.replica = replica_idx
logger.info((
'generating - ensemble `%s` found sample '
'replica %d, length %d\n')
% (
ens.name, sample.replica, len(sample)
))
self.append(sample)
if reuse_strategy != 'all':
# we mark the trajectory and its reversed as used
if sample.trajectory not in used_trajectories and (
not reuse_strategy.endswith('symmetric') or
sample.trajectory.reversed in used_trajectories
):
used_trajectories.append(sample.trajectory)
# if reuse_strategy.endswith('symmetric'):
# used_trajectories.append(
# sample.trajectory.reversed)
# we want the list of used_trajectories to be
# sorted. Short ones first. So if we have to chose
# from the used_ones, use the shortest one
# used_trajectories = sorted(
# used_trajectories, key=len)
# found a sample in this category so remove it for
# other tries
del ensembles_to_fill[idx]
# do not try other ensembles in this category
break
found_samples_str = \
found_samples_str[:pos] + \
(str(str_idx + 1)[0] if found else '.') + \
found_samples_str[pos + 1:]
refresh_output((
'# finished generating: still missing %d samples\n'
'%s\n'
) % (
len(ensembles_to_fill),
found_samples_str
), ipynb_display_only=True, print_anyway=False)
return self
def initial_conditions_from_trajectories(self, trajectories,
sample_set=None,
strategies=None,
preconditions=None,
reuse_strategy='avoid-symmetric',
engine=None):
"""
Create a SampleSet with as many initial samples as possible.
The goal of this is to give the initial SampleSet that would be
desired.
Parameters
----------
trajectories : list of :class:`.Trajectory` or :class:`.Trajectory`
the input trajectories to use
sample_set : :class:`.SampleSet`, optional
if given, add samples to this sampleset. Default is None, which
means that this will start a new sampleset.
strategies : dict
a dict that specifies the options used when ensemble functions
are used to create a new sample.
preconditions : list of str
a list of possible steps to modify the initial list of trajectories.
possible choices are
1. `sort-shortest` - sorting by shortest first,
2. `sort_median` - sorting by the middle one first and then in
move away from the median length
3. `sort-longest` - sorting by the longest first
4. `reverse` - reverse the order and
5. `mirror` which will add the reversed trajectories to the
list in the same order
Default is `None` which means to do nothing.
reuse_strategy : str
if `avoid` then reusing the same same trajectory twice is avoided.
`avoid-symmetric` will also remove reversed copies
if possible. `all` will not attempt to avoid already existing ones.
`once` will strictly not reuse a trajectory and `once-symmetric`
will also not use reversed copies.
engine : :class:`openpathsampling.engines.DyanmicsEngine`
the engine used for extending moves
Returns
-------
:class:`.SampleSet`
sampleset with samples for every initial ensemble for this
scheme that could be satisfied by the given trajectories
See Also
--------
list_initial_ensembles
check_initial_conditions
assert_initial_conditions
"""
if sample_set is None:
sample_set = paths.SampleSet([])
ensembles = self.list_initial_ensembles()
sample_set = sample_set.generate_from_trajectories(
ensembles,
trajectories,
preconditions,
strategies,
reuse_strategy,
engine
)
refresh_output(self.initial_conditions_report(sample_set),
ipynb_display_only=True, print_anyway=False)
return sample_set
def initial_conditions_from_trajectories(self, trajectories,
sample_set=None,
strategies=None,
preconditions=None,
reuse_strategy='avoid-symmetric',
engine=None):
"""
Create a SampleSet with as many initial samples as possible.
The goal of this is to give the initial SampleSet that would be
desired.
Parameters
----------
trajectories : list of :class:`.Trajectory` or :class:`.Trajectory`
the input trajectories to use
sample_set : :class:`.SampleSet`, optional
if given, add samples to this sampleset. Default is None, which
means that this will start a new sampleset.
strategies : dict
a dict that specifies the options used when ensemble functions
are used to create a new sample.
preconditions : list of str
a list of possible steps to modify the initial list of trajectories.
possible choices are
1. `sort-shortest` - sorting by shortest first,
2. `sort_median` - sorting by the middle one first and then in
move away from the median length
3. `sort-longest` - sorting by the longest first
4. `reverse` - reverse the order and
5. `mirror` which will add the reversed trajectories to the
list in the same order
Default is `None` which means to do nothing.
reuse_strategy : str
if `avoid` then reusing the same same trajectory twice is avoided.
`avoid-symmetric` will also remove reversed copies
if possible. `all` will not attempt to avoid already existing ones.
`once` will strictly not reuse a trajectory and `once-symmetric`
will also not use reversed copies.
engine : :class:`openpathsampling.engines.DyanmicsEngine`
the engine used for extending moves
Returns
-------
:class:`.SampleSet`
sampleset with samples for every initial ensemble for this
scheme that could be satisfied by the given trajectories
See Also
--------
list_initial_ensembles
check_initial_conditions
assert_initial_conditions
"""
if sample_set is None:
sample_set = paths.SampleSet([])
ensembles = self.list_initial_ensembles()
sample_set = sample_set.generate_from_trajectories(
ensembles,
trajectories,
preconditions,
strategies,
reuse_strategy,
engine
)
refresh_output(self.initial_conditions_report(sample_set),
ipynb_display_only=True, print_anyway=False)
return sample_set
def generate_from_trajectories(self,
ensembles,
trajectories,
preconditions=None,
strategies=None,
reuse_strategy='avoid',
engine=None):
"""
Create a SampleSet with as many initial samples as possible.
The goal of this is to give the initial SampleSet that would be
desired.
Parameters
----------
trajectories : list of :class:`.Trajectory` or :class:`.Trajectory`
the input trajectories to use
ensembles : list of :class:`Ensemble` or list of :class:`Ensemble`
the list of ensembles to be generated. If an element is itself a
list then one sample for one of the ensembles in that list if
generated
preconditions : list of str
a list of possible steps to modify the initial list of trajectories.
possible choices are
1. `sort-shortest` - sorting by shortest first,
2. `sort_median` - sorting by the middle one first and then in
move away from the median length
3. `sort-longest` - sorting by the longest first
4. `reverse` - reverse the order and
5. `mirror` which will add the reversed trajectories to the
list in the same order
Default is `None` which means to do nothing.
strategies : dict
a dict that specifies the options used when ensemble functions
are used to create a new sample.
reuse_strategy : str
if `avoid` then in a second attempt the used trajectories are
tried
engine : :class:`openpathsampling.engines.DyanmicsEngine`
the engine used for extending moves
Returns
-------
:class:`.SampleSet`
sampleset with samples for every initial ensemble for this
scheme that could be satisfied by the given trajectories
See Also
--------
list_initial_ensembles
"""
implemented_strategies = [
'get', # look for existing trajectories
'split', # look for existing sub-trajectories
'extend-complex', # try to extend long sub-trajectories
'extend-minimal' # try to extend short sub-trajectories
]
implemented_preconditions = [
'sort-shortest', 'sort-median', 'sort-longest', 'reverse', 'mirror'
]
if preconditions is None:
preconditions = ['mirror']
# create a list of trajectories
trajectories = paths.Trajectory._to_list_of_trajectories(trajectories)
for pre in preconditions:
if pre not in implemented_preconditions:
raise RuntimeError(
'%s is not a valid precondition strategy. Choose from %s.'
% (pre, implemented_preconditions))
if pre == 'sort-shortest':
trajectories = sorted(trajectories, key=len)
elif pre == 'sort-longest':
trajectories = sorted(trajectories, key=len)
elif pre == 'sort-median':
sorted_trajectories = sorted(trajectories, key=len)
trajectories = list([
p for p2 in zip(
sorted_trajectories[len(sorted_trajectories) / 2:],
reversed(
sorted_trajectories[:len(sorted_trajectories) /
2])) for p in p2
])
if len(sorted_trajectories) & 1:
trajectories.append(sorted_trajectories[-1])
elif pre == 'reverse':
trajectories = list(reversed(trajectories))
elif pre == 'mirror':
trajectories = trajectories + \
[traj.reversed for traj in trajectories]
# let's always try the short trajectories first
# print map(lambda x: hex(id(x)), trajectories)
# we will try forward/backward interleaved
used_trajectories = []
# if we start with an existing sample set look at what we got
# if we avoid we move the used ones to the back of the list
# if we remove we remove the used ones
for s in self:
traj = s.trajectory
if traj in trajectories:
used_trajectories.append(traj)
used_trajectories = sorted(used_trajectories, key=len)
# print map(lambda x: hex(id(x)), used_trajectories)
ensembles = [[x] if type(x) is not list else x for x in ensembles]
# 1. look in the existing sample_set
ensembles_to_fill, extra_ensembles = self.check_ensembles(ensembles)
# we reverse because we want to be able to remove elements
# from the list as we discover samples. This is easier to do
# when the list is traversed backwards since indices to not
# change, hence we reverse the list of ensemble and then traverse it
# in reversed order
ensembles_to_fill = \
list(reversed(ensembles_to_fill))
# 2. try strategies
if strategies is None:
# this is the default
strategies = ['get', 'split']
for idx, strategy in enumerate(strategies):
if type(strategy) is str:
strategies[idx] = (strategy, dict())
if strategies[idx][0] not in implemented_strategies:
raise RuntimeError(
'Strategy `%s` is not known. Chose from %s.' %
(strategies[idx][0], implemented_strategies))
found_samples_str = ''
for pos, ens_list in enumerate(ensembles):
found_samples_str += '.' if ens_list in ensembles_to_fill else '+'
for str_idx, (strategy, options) in enumerate(strategies):
for idx, ens_list in reversed(list(enumerate(ensembles_to_fill))):
pos = ensembles.index(ens_list)
found_samples_str = \
found_samples_str[:pos] + \
'?' + found_samples_str[pos + 1:]
refresh_output(
('# trying strategy #%d `%s`: still missing %d samples\n'
'%s\n') % (str_idx + 1, strategy, len(ensembles_to_fill),
found_samples_str),
ipynb_display_only=True,
print_anyway=False)
if type(ens_list) is not list:
ens_list = [ens_list]
found = False
for ens in ens_list:
# create the list of options to be passed on
opts = {
key: value
for key, value in options.items()
if key not in ['exclude']
}
# exclude contains the Ensemble classes to be ignored
if 'exclude' in options:
if isinstance(ens, options['exclude']):
continue
# fill only the first in ens_list that can be filled
sample = None
if strategy == 'get':
sample = ens.get_sample_from_trajectories(
trajectories=trajectories,
used_trajectories=used_trajectories,
reuse_strategy=reuse_strategy,
**opts)
elif strategy == 'split':
sample = ens.split_sample_from_trajectories(
trajectories=trajectories,
used_trajectories=used_trajectories,
reuse_strategy=reuse_strategy,
**opts)
elif strategy == 'extend-complex' and engine:
if hasattr(ens, 'extend_sample_from_trajectories'):
sample = ens.extend_sample_from_trajectories(
trajectories=trajectories,
engine=engine,
level='complex',
**opts)
elif strategy == 'extend-minimal' and engine:
if hasattr(ens, 'extend_sample_from_trajectories'):
sample = ens.extend_sample_from_trajectories(
trajectories=trajectories,
engine=engine,
level='minimal',
**opts)
elif strategy == 'extend-native' and engine:
if hasattr(ens, 'extend_sample_from_trajectories'):
sample = ens.extend_sample_from_trajectories(
trajectories=trajectories,
engine=engine,
level='native',
**opts)
# now, if we've found a sample, add it and
# make sure we chose a proper replica ID
if sample is not None:
found = True
# another way would be to look for the smallest not
# taken id. This one is simpler
if len(self.replicas) > 0:
replica_idx = max(0, max(self.replicas) + 1)
else:
replica_idx = 0
sample.replica = replica_idx
logger.info(('generating - ensemble `%s` found sample '
'replica %d, length %d\n') %
(ens.name, sample.replica, len(sample)))
self.append(sample)
if reuse_strategy != 'all':
# we mark the trajectory and its reversed as used
if sample.trajectory not in used_trajectories and (
not reuse_strategy.endswith('symmetric')
or sample.trajectory.reversed
in used_trajectories):
used_trajectories.append(sample.trajectory)
# if reuse_strategy.endswith('symmetric'):
# used_trajectories.append(
# sample.trajectory.reversed)
# we want the list of used_trajectories to be
# sorted. Short ones first. So if we have to chose
# from the used_ones, use the shortest one
# used_trajectories = sorted(
# used_trajectories, key=len)
# found a sample in this category so remove it for
# other tries
del ensembles_to_fill[idx]
# do not try other ensembles in this category
break
found_samples_str = \
found_samples_str[:pos] + \
(str(str_idx + 1)[0] if found else '.') + \
found_samples_str[pos + 1:]
refresh_output(('# finished generating: still missing %d samples\n'
'%s\n') % (len(ensembles_to_fill), found_samples_str),
ipynb_display_only=True,
print_anyway=False)
return self
