def write_all_trajectories(self,
input_dir,
output_dir,
stride,
max_rmsd,
min_gens,
center_conformations,
num_proc,
input_style,
update=False):
"""
Convert all of the trajectories in the FAH project in input_dir to
lh5 trajectory files which will be placed in output dir.
If the 'update' flag is set, then will use the memory object to check for
previously converted data, and add to it (rather than reconverting everything).
This functionality can be more cleanly called through the update_trajectories()
method.
Parameters
----------
input_dir : str
The directory to look for XTC/DCD files in.
output_dir : str
The place to write the converted lh5s
stride : int
The size of the stride to employ. E.g., if stride = 3, the script
keeps every 3rd MD snapshot from the original data. Useful to throw
away highly correlated data if snapshots were saved frequently.
max_rmsd : float
Throw away any data that is further than `max_rmsd` (in nm) from the
pdb file associated with the project. This is used as a sanity check
to prevent including, e.g. data from a simulation that is blowing up.
min_gens : int
Discard trajectories with fewer than `min_gens` generations.
center_conformations : bool
Whether to center the converted (lh5) conformations.
num_proc : int
Number of processors to employ. Note that this function is typically
I/O limited, so paralellism is unlikely to yield much gain.
input_style : {'FAH', 'FILE'}
If you use input_style = 'FAH', this code uses knowledge of the
RUN*/CLONE* directory structure to yield all the CLONE directories.
If you use input_style = 'FILE', this code uses os.walk() which is
A LOT slower because it has to stat every file, but is capable of
recursively searching for xtc files to arbitrary depths.
update : bool
If `True`, then tries to figure out what data has already been converted
by reading the "memory state" in the provided ProjectInfo file, and only
converts new data. If `False`, does a fresh re-convert.
Notes
-----
Since sometimes a conversion fails, we collect all trajectories at the
end and renumber them such that they are contiguously numbered.
"""
if update:
assert os.path.exists(output_dir)
else:
try:
os.mkdir(output_dir)
except OSError:
logger.error('Error: The directory %s already exists',
output_dir)
sys.exit(1)
intermediate_filename_root = '_trj' # A placeholder name
#dtm does not play nice with OpenMP
use_parallel_rmsd = (num_proc != 'use_dtm_instead')
jobs = []
for i, clone_dir in enumerate(
self.yield_xtc_directories(input_dir, input_style)):
job = {
'clone_dir': clone_dir,
'output_dir': output_dir,
'pdb_file': self.pdb_topology,
'trajectory_number': i,
'stride': stride,
'max_rmsd': max_rmsd,
'min_gens': min_gens,
'center_conformations': center_conformations,
'memory_check': update,
'omp_parallel_rmsd': use_parallel_rmsd
}
jobs.append(job)
if len(jobs) == 0:
raise RuntimeError('No conversion jobs found!')
if num_proc == 'use_dtm_instead':
# use DTM mpi parallel map
dtm.map(self.write_trajectory_mapper, jobs)
elif num_proc > 1:
# use multiprocessing
pool = Pool(processes=num_proc)
pool.map(self.write_trajectory_mapper, jobs)
else:
# use regular serial execution
map(self.write_trajectory_mapper, jobs)
# Rename trajectory files such that they have contiguous numbering
logger.info(
"Finished Generating Trajectories. Renaming them now in contiguous order"
)
mapping = {
} # document the directory changes, allowing us to update memory
for i, filename in enumerate(sorted(os.listdir(output_dir),
key=keynat)):
path = os.path.join(output_dir, filename)
new_path = os.path.join(output_dir, "trj%d.lh5" % i)
os.rename(path, new_path)
mapping[path] = new_path
# update the memory hash to accound for our renumbering
for key in self.memory.keys():
if key not in ['convert_parameters', 'SerializerFilename']:
logger.info("%s --> %s", self.memory[key][0],
mapping[self.memory[key][0]])
self.memory[key][0] = mapping[self.memory[key][0]]
# save the parameters used for this run in the memory file, and write to disk
logger.info("Generating Project File: %s", self.projectinfo_file)
if update:
try:
os.remove(self.projectinfo_file
) # if we are updating, just start w fresh slate
except:
pass
self.memory['convert_parameters'] = (input_dir, output_dir, stride,
max_rmsd, min_gens,
center_conformations, num_proc,
self.projectinfo_file,
input_style)
Project.CreateProjectFromDir(Filename=self.projectinfo_file,
TrajFilePath=output_dir,
TrajFileBaseName='trj',
TrajFileType='.lh5',
ConfFilename=self.pdb_topology,
initial_memory=cPickle.dumps(self.memory))
logger.info("Data converted properly.")
return
