def load(cls, filename, clear_session=True):
""" Loads parameters into model. Careful: this clears the whole TF session!!
"""
from deep_boltzmann.util import load_obj
if clear_session:
keras.backend.clear_session()
D = load_obj(filename)
prior = D['prior']
layerdicts = D['layers']
layers = [eval(d['type']).from_dict(d) for d in layerdicts]
return InvNet(D['dim'], layers, prior=prior)
def load(cls, filename):
""" Loads parameters into model. The resulting model is just a data container.
"""
from deep_boltzmann.util import load_obj
D = load_obj(filename)
umbrellas = [UmbrellaModel.from_dict(u) for u in D['umbrellas']]
us = cls(None,
None,
None,
None,
len(umbrellas),
umbrellas[0].k_umbrella,
umbrellas[0].m_umbrella,
umbrellas[-1].m_umbrella,
forward_backward=D['forward_backward'])
us.umbrellas = umbrellas
if 'rc_discretization' in D:
us.rc_discretization = D['rc_discretization']
us.rc_free_energies = D['rc_free_energies']
return us
mm_cyc9 = MM(toppar_cyc9, align=0.0)
top_cyc9 = toppar_cyc9.mdtraj_topology()
rotamer_mapper = RotamerMapper(9)
# get C torsions
Ctor = []
for i, tor in enumerate(toppar_cyc9.torsion_indices):
is_Ctor = True
for j in tor:
if not toppar_cyc9.atom_names[j].startswith('C'):
is_Ctor = False
if is_Ctor:
Ctor.append(i)
# load data
remd_dict = load_obj('../local_data/out/hydrocarbon_cyc9/remd_data.pkl')
traj = mdtraj.Trajectory(
remd_dict['trajs'][0].reshape(remd_dict['trajs'][0].shape[0],
toppar_cyc9.natoms, 3), top_cyc9)
traj = traj[10000:14000]
traj = traj.superpose(traj[0])
xtrain = traj.xyz.reshape(
(traj.xyz.shape[0], traj.xyz.shape[1] * traj.xyz.shape[2]))
network_resample = create_RealNVPNet(mm_cyc9, nlayers=4, nl_activation='tanh')
resample(mm_cyc9,
Ctor,
network_resample,
xtrain,
traj[0],
epochsML=300,