def energy(self):
xyz = utils.get_simulation_xyz(self._simulation)
snapshot = mdtraj.Trajectory(xyz, self.topology)
self._prior_energy = self._simulation.context.getState(getEnergy=True).getPotentialEnergy()._value
self._potential_energy = self.potential(snapshot)
self._total_energy = self._potential_energy + self._prior_energy
return self._total_energy
def energy(self):
xyz = utils.get_simulation_xyz(self._simulation)
snapshot = mdtraj.Trajectory(xyz, self.topology)
self._prior_energy = self._simulation.context.getState(
getEnergy=True).getPotentialEnergy()._value
self._potential_energy = self.potential(snapshot)
self._total_energy = self._potential_energy + self._prior_energy
return self._total_energy
def sample(self, num_moves, output_target):
"""
Run the sampler for `num_moves`.
Parameteres
-----------
num_moves : int
The number of moves to generate.
output_target : file
A file object that serialized results will be sent to. Could be
a pipe or something similar.
Returns
-------
sample : mdtraj.trajectory
An `num_moves` length trajectory, sampled from the potential
assoicated with the
"""
# need to figure out how we're going to deal with the traj data -- cant
# keep it all in memory. Would be good to dump it into a single h5 db
reporter = reporters.HDF5Reporter(output_target,
self.mc_length_increment,
coordinates=True, time=False, cell=False,
potentialEnergy=False, kineticEnergy=False,
temperature=False, velocities=False,
atomSubset=None)
self._simulation.reporters.append(reporter)
# perform the actual monte carlo
current_energy = self.energy
logger.debug('PRIOR POTNT NEW CRRNT ACCPT RATIO')
for mi in range(num_moves):
# step forward in time
self.total_moves_attempted += 1
old_pos = self._simulation.context.getState(getPositions=True).getPositions()
self._simulation.step(self.steps_per_iter)
# accept or reject according to Metropolis
new_energy = self.energy
# accept
if (new_energy < current_energy):
accept_str = 'Y-lwr' # move accepted, new energy lower
self.accepted += 1
self.positions = utils.get_simulation_xyz(self._simulation)
current_energy = new_energy
elif (np.random.rand() < np.exp( (current_energy - new_energy) / (k_boltz * self.temperature))):
accept_str = 'Y-rnd' # move accepted due to random chance
self.accepted += 1
self.positions = utils.get_simulation_xyz(self._simulation)
current_energy = new_energy
# reject
else:
accept_str = 'N'
self._simulation.context.setPositions(old_pos)
# perform adaptive update on the number of steps performed per attempt
# to try and get to the `target_accept_percent` acceptance ratio
self.accepted_ratio = float(self.accepted) / float(self.total_moves_attempted)
if self.accepted_ratio > self.target_accept_percent:
self.steps_per_iter /= self.scaling_speed
else:
self.steps_per_iter *= self.scaling_speed
self.steps_per_iter = int(self.steps_per_iter)
self.steps_per_iter -= self.steps_per_iter % self.mc_length_increment
self.steps_per_iter = max(self.steps_per_iter, self.mc_length_increment)
report_str = '%.2e ' * 4 + '%s %.2f'
logger.debug(report_str % (self._prior_energy,
self._potential_energy,
new_energy,
current_energy,
accept_str,
self.accepted_ratio))
# close repoter
reporter.close()
return
def sample(self, num_moves, output_target):
"""
Run the sampler for `num_moves`.
Parameteres
-----------
num_moves : int
The number of moves to generate.
output_target : file
A file object that serialized results will be sent to. Could be
a pipe or something similar.
Returns
-------
sample : mdtraj.trajectory
An `num_moves` length trajectory, sampled from the potential
assoicated with the
"""
# need to figure out how we're going to deal with the traj data -- cant
# keep it all in memory. Would be good to dump it into a single h5 db
reporter = reporters.HDF5Reporter(output_target,
self.mc_length_increment,
coordinates=True,
time=False,
cell=False,
potentialEnergy=False,
kineticEnergy=False,
temperature=False,
velocities=False,
atomSubset=None)
self._simulation.reporters.append(reporter)
# perform the actual monte carlo
current_energy = self.energy
logger.debug('PRIOR POTNT NEW CRRNT ACCPT RATIO')
for mi in range(num_moves):
# step forward in time
self.total_moves_attempted += 1
old_pos = self._simulation.context.getState(
getPositions=True).getPositions()
self._simulation.step(self.steps_per_iter)
# accept or reject according to Metropolis
new_energy = self.energy
# accept
if (new_energy < current_energy):
accept_str = 'Y-lwr' # move accepted, new energy lower
self.accepted += 1
self.positions = utils.get_simulation_xyz(self._simulation)
current_energy = new_energy
elif (np.random.rand() < np.exp(
(current_energy - new_energy) / (k_boltz * self.temperature))):
accept_str = 'Y-rnd' # move accepted due to random chance
self.accepted += 1
self.positions = utils.get_simulation_xyz(self._simulation)
current_energy = new_energy
# reject
else:
accept_str = 'N'
self._simulation.context.setPositions(old_pos)
# perform adaptive update on the number of steps performed per attempt
# to try and get to the `target_accept_percent` acceptance ratio
self.accepted_ratio = float(self.accepted) / float(
self.total_moves_attempted)
if self.accepted_ratio > self.target_accept_percent:
self.steps_per_iter /= self.scaling_speed
else:
self.steps_per_iter *= self.scaling_speed
self.steps_per_iter = int(self.steps_per_iter)
self.steps_per_iter -= self.steps_per_iter % self.mc_length_increment
self.steps_per_iter = max(self.steps_per_iter,
self.mc_length_increment)
report_str = '%.2e ' * 4 + '%s %.2f'
logger.debug(
report_str %
(self._prior_energy, self._potential_energy, new_energy,
current_energy, accept_str, self.accepted_ratio))
# close repoter
reporter.close()
return
