def iterload(filename, chunk=100, **kwargs):
"""An iterator over a trajectory from one or more files on disk, in fragments
This may be more memory efficient than loading an entire trajectory at
once
Parameters
----------
filename : str
Path to the trajectory file on disk
chunk : int
Number of frames to load at once from disk per iteration. If 0, load all.
Other Parameters
----------------
top : {str, Trajectory, Topology}
Most trajectory formats do not contain topology information. Pass in
either the path to a RCSB PDB file, a trajectory, or a topology to
supply this information. This option is not required for the .h5, .lh5,
and .pdb formats, which already contain topology information.
stride : int, default=None
Only read every stride-th frame.
atom_indices : array_like, optional
If not none, then read only a subset of the atoms coordinates from the
file. This may be slightly slower than the standard read because it
requires an extra copy, but will save memory.
See Also
--------
load, load_frame
Examples
--------
>>> import mdtraj as md
>>> for chunk in md.iterload('output.xtc', top='topology.pdb')
>>> print chunk
<mdtraj.Trajectory with 100 frames, 423 atoms at 0x110740a90>
<mdtraj.Trajectory with 100 frames, 423 atoms at 0x110740a90>
<mdtraj.Trajectory with 100 frames, 423 atoms at 0x110740a90>
<mdtraj.Trajectory with 100 frames, 423 atoms at 0x110740a90>
<mdtraj.Trajectory with 100 frames, 423 atoms at 0x110740a90>
"""
stride = kwargs.get('stride', 1)
atom_indices = cast_indices(kwargs.get('atom_indices', None))
if chunk % stride != 0 and filename.endswith('.dcd'):
raise ValueError('Stride must be a divisor of chunk. stride=%d does not go '
'evenly into chunk=%d' % (stride, chunk))
if chunk == 0:
yield load(filename, **kwargs)
# If chunk was 0 then we want to avoid filetype-specific code in case of undefined behavior in various file parsers.
else:
skip = kwargs.get('skip', 0)
if filename.endswith('.h5'):
if 'top' in kwargs:
warnings.warn('top= kwarg ignored since file contains topology information')
with HDF5TrajectoryFile(filename) as f:
if skip > 0:
xyz, _, _, _ = f.read(skip, atom_indices=atom_indices)
if len(xyz) == 0:
raise StopIteration()
if atom_indices is None:
topology = f.topology
else:
topology = f.topology.subset(atom_indices)
while True:
data = f.read(chunk*stride, stride=stride, atom_indices=atom_indices)
if data == []:
raise StopIteration()
in_units_of(data.coordinates, f.distance_unit, Trajectory._distance_unit, inplace=True)
in_units_of(data.cell_lengths, f.distance_unit, Trajectory._distance_unit, inplace=True)
yield Trajectory(xyz=data.coordinates, topology=topology,
time=data.time, unitcell_lengths=data.cell_lengths,
unitcell_angles=data.cell_angles)
if filename.endswith('.lh5'):
if 'top' in kwargs:
warnings.warn('top= kwarg ignored since file contains topology information')
with LH5TrajectoryFile(filename) as f:
if atom_indices is None:
topology = f.topology
else:
topology = f.topology.subset(atom_indices)
ptr = 0
if skip > 0:
xyz, _, _, _ = f.read(skip, atom_indices=atom_indices)
if len(xyz) == 0:
raise StopIteration()
while True:
xyz = f.read(chunk*stride, stride=stride, atom_indices=atom_indices)
if len(xyz) == 0:
raise StopIteration()
in_units_of(xyz, f.distance_unit, Trajectory._distance_unit, inplace=True)
time = np.arange(ptr, ptr+len(xyz)*stride, stride)
ptr += len(xyz)*stride
yield Trajectory(xyz=xyz, topology=topology, time=time)
elif filename.endswith('.xtc'):
topology = _parse_topology(kwargs.get('top', None))
with XTCTrajectoryFile(filename) as f:
if skip > 0:
xyz, _, _, _ = f.read(skip)
if len(xyz) == 0:
raise StopIteration()
while True:
xyz, time, step, box = f.read(chunk*stride, stride=stride, atom_indices=atom_indices)
if len(xyz) == 0:
raise StopIteration()
in_units_of(xyz, f.distance_unit, Trajectory._distance_unit, inplace=True)
in_units_of(box, f.distance_unit, Trajectory._distance_unit, inplace=True)
trajectory = Trajectory(xyz=xyz, topology=topology, time=time)
trajectory.unitcell_vectors = box
yield trajectory
elif filename.endswith('.dcd'):
topology = _parse_topology(kwargs.get('top', None))
with DCDTrajectoryFile(filename) as f:
ptr = 0
if skip > 0:
xyz, _, _ = f.read(skip, atom_indices=atom_indices)
if len(xyz) == 0:
raise StopIteration()
while True:
# for reasons that I have not investigated, dcdtrajectory file chunk and stride
# together work like this method, but HDF5/XTC do not.
xyz, box_length, box_angle = f.read(chunk, stride=stride, atom_indices=atom_indices)
if len(xyz) == 0:
raise StopIteration()
in_units_of(xyz, f.distance_unit, Trajectory._distance_unit, inplace=True)
in_units_of(box_length, f.distance_unit, Trajectory._distance_unit, inplace=True)
time = np.arange(ptr, ptr+len(xyz)*stride, stride)
ptr += len(xyz)*stride
yield Trajectory(xyz=xyz, topology=topology, time=time, unitcell_lengths=box_length,
unitcell_angles=box_angle)
else:
log.critical("loading complete traj into mem! This might no be desired.")
t = load(filename, **kwargs)
for i in range(skip, len(t), chunk):
yield t[i:i+chunk]
def __init__(self, filename, chunk=1000, **kwargs):
"""An iterator over a trajectory from one or more files on disk, in fragments
This may be more memory efficient than loading an entire trajectory at
once
Parameters
----------
filename : str
Path to the trajectory file on disk
chunk : int
Number of frames to load at once from disk per iteration. If 0, load all.
Other Parameters
----------------
top : {str, Trajectory, Topology}
Most trajectory formats do not contain topology information. Pass in
either the path to a RCSB PDB file, a trajectory, or a topology to
supply this information. This option is not required for the .h5, .lh5,
and .pdb formats, which already contain topology information.
stride : int, default=None
Only read every stride-th frame.
atom_indices : array_like, optional
If not none, then read only a subset of the atoms coordinates from the
file. This may be slightly slower than the standard read because it
requires an extra copy, but will save memory.
See Also
--------
load, load_frame
Examples
--------
>>> import mdtraj as md
>>> for chunk in md.iterload('output.xtc', top='topology.pdb') # doctest: +SKIP
... print chunk # doctest: +SKIP
<mdtraj.Trajectory with 100 frames, 423 atoms at 0x110740a90>
<mdtraj.Trajectory with 100 frames, 423 atoms at 0x110740a90>
<mdtraj.Trajectory with 100 frames, 423 atoms at 0x110740a90>
<mdtraj.Trajectory with 100 frames, 423 atoms at 0x110740a90>
<mdtraj.Trajectory with 100 frames, 423 atoms at 0x110740a90>
"""
self._filename = filename
self._stride = kwargs.pop('stride', 1)
self._atom_indices = cast_indices(kwargs.pop('atom_indices', None))
self._top = kwargs.pop('top', None)
self._skip = kwargs.pop('skip', 0)
self._kwargs = kwargs
self._chunksize = chunk
self._extension = _get_extension(self._filename)
self._closed = False
self._seeked = False
if self._extension not in _TOPOLOGY_EXTS:
self._topology = load_topology_cached(self._top)
else:
self._topology = self._top
if self._extension in ('pdb', 'pdb.gz'):
raise Exception("Not supported as trajectory format {ext}".format(
ext=self._extension))
self._mode = None
if isinstance(self._stride, np.ndarray):
self._mode = 'random_access'
self._f = (lambda x: md_open(x, n_atoms=self._topology.n_atoms)
if self._extension in
('.crd', '.mdcrd') else md_open(self._filename))(
self._filename)
self._ra_it = self._random_access_generator(self._f)
else:
self._mode = 'traj'
self._f = (lambda x: md_open(x, n_atoms=self._topology.n_atoms)
if self._extension in
('.crd', '.mdcrd') else md_open(self._filename))(
self._filename)
# offset array handling
offsets = kwargs.pop('offsets', None)
if hasattr(self._f, 'offsets') and offsets is not None:
self._f.offsets = offsets
def iterload(filename, chunk=100, **kwargs):
"""An iterator over a trajectory from one or more files on disk, in fragments
This may be more memory efficient than loading an entire trajectory at
once
Parameters
----------
filename : str
Path to the trajectory file on disk
chunk : int
Number of frames to load at once from disk per iteration. If 0, load all.
Other Parameters
----------------
top : {str, Trajectory, Topology}
Most trajectory formats do not contain topology information. Pass in
either the path to a RCSB PDB file, a trajectory, or a topology to
supply this information. This option is not required for the .h5, .lh5,
and .pdb formats, which already contain topology information.
stride : int, default=None
Only read every stride-th frame.
atom_indices : array_like, optional
If not none, then read only a subset of the atoms coordinates from the
file. This may be slightly slower than the standard read because it
requires an extra copy, but will save memory.
See Also
--------
load, load_frame
Examples
--------
>>> import mdtraj as md
>>> for chunk in md.iterload('output.xtc', top='topology.pdb') # doctest: +SKIP
... print chunk # doctest: +SKIP
<mdtraj.Trajectory with 100 frames, 423 atoms at 0x110740a90>
<mdtraj.Trajectory with 100 frames, 423 atoms at 0x110740a90>
<mdtraj.Trajectory with 100 frames, 423 atoms at 0x110740a90>
<mdtraj.Trajectory with 100 frames, 423 atoms at 0x110740a90>
<mdtraj.Trajectory with 100 frames, 423 atoms at 0x110740a90>
"""
stride = kwargs.pop('stride', 1)
atom_indices = cast_indices(kwargs.pop('atom_indices', None))
top = kwargs.pop('top', None)
skip = kwargs.pop('skip', 0)
extension = _get_extension(filename)
if extension not in _TOPOLOGY_EXTS:
topology = _parse_topology(top)
else:
topology = top
if chunk == 0:
# If chunk was 0 then we want to avoid filetype-specific code
# in case of undefined behavior in various file parsers.
# TODO: this will first apply stride, then skip!
if extension not in _TOPOLOGY_EXTS:
kwargs['top'] = top
yield load(filename, **kwargs)[skip:]
elif extension in ('.pdb', '.pdb.gz'):
# the PDBTrajectortFile class doesn't follow the standard API. Fixing it
# to support iterload could be worthwhile, but requires a deep refactor.
t = load(filename, stride=stride, atom_indices=atom_indices)
for i in range(0, len(t), chunk):
yield t[i:i + chunk]
elif isinstance(stride, np.ndarray):
with (lambda x: open(x, n_atoms=topology.n_atoms) if extension in
('.crd', '.mdcrd') else open(filename))(filename) as f:
x_prev = 0
curr_size = 0
traj = []
leftovers = []
for k, g in groupby(enumerate(stride), lambda a: a[0] - a[1]):
grouped_stride = list(map(itemgetter(1), g))
seek_offset = (1 if x_prev != 0 else 0)
seek_to = grouped_stride[0] - x_prev - seek_offset
f.seek(seek_to, whence=1)
x_prev = grouped_stride[-1]
group_size = len(grouped_stride)
if curr_size + group_size > chunk:
leftovers = grouped_stride
else:
local_traj = _get_local_traj_object(
atom_indices, extension, f, group_size, topology,
**kwargs)
traj.append(local_traj)
curr_size += len(grouped_stride)
if curr_size == chunk:
yield _efficient_traj_join(traj)
curr_size = 0
traj = []
while leftovers:
local_chunk = leftovers[:min(chunk, len(leftovers))]
local_traj = _get_local_traj_object(
atom_indices, extension, f, len(local_chunk), topology,
**kwargs)
traj.append(local_traj)
leftovers = leftovers[min(chunk, len(leftovers)):]
curr_size += len(local_chunk)
if curr_size == chunk:
yield _efficient_traj_join(traj)
curr_size = 0
traj = []
if traj:
yield _efficient_traj_join(traj)
raise StopIteration()
else:
with (lambda x: open(x, n_atoms=topology.n_atoms) if extension in
('.crd', '.mdcrd') else open(filename))(filename) as f:
if skip > 0:
f.seek(skip)
while True:
if extension not in _TOPOLOGY_EXTS:
traj = f.read_as_traj(topology,
n_frames=chunk * stride,
stride=stride,
atom_indices=atom_indices,
**kwargs)
else:
traj = f.read_as_traj(n_frames=chunk * stride,
stride=stride,
atom_indices=atom_indices,
**kwargs)
if len(traj) == 0:
raise StopIteration()
yield traj
