def __init__(self, architecture):
super(NormA2MDDensity, self).__init__()
# self.normalize = normalize
try:
common_net_architecture = architecture['common_net']
atom_net_architecture = architecture['atom_net']
bond_net_architecture = architecture['bond_net']
subnet = architecture['subnet']
except KeyError:
raise IOError("architecture dict did not contain the right fields")
self.common_net = modules.TorchElementSpecificA2MDNN(
nodes=common_net_architecture,
elements=[1, 6, 7, 8],
device=DEVICE)
self.atom_net = modules.TorchElementSpecificA2MDNN(
nodes=atom_net_architecture, elements=[1, 6, 7, 8], device=DEVICE)
self.bond_net = modules.TorchPairSpecificA2MDNN(
nodes=bond_net_architecture, elements=[1, 6, 7, 8])
self.feature_extractor = modules.TorchaniFeats(net=subnet,
device=DEVICE)
for param in self.feature_extractor.parameters():
param.requires_grad = False
self.normalizer = modules.QPChargeNormalization()
def __init__(self, architecture, device):
from a2mdnet.a2mdt.modules import A2MDt
super(MultTaskDensityCoupled, self).__init__()
self.device = device
try:
common_net_architecture = architecture['common_net']
atom_net_architecture = architecture['atom_net']
mol_net_architecture = architecture['mol_net']
bond_net_architecture = architecture['bond_net']
feature_extraction_layer = architecture['fe_layer']
feature_extraction_net = architecture['fe_net']
density_function_iso_order = architecture['iso_order']
density_function_aniso_order = architecture['aniso_order']
coefficients_distribution = architecture[
'coefficients_distribution']
except KeyError:
raise IOError("architecture dict did not contain the right fields")
self.common_net = modules.TorchElementSpecificA2MDNN(
nodes=common_net_architecture,
elements=[1, 6, 7, 8],
device=self.device)
self.mol_net = modules.TorchElementSpecificA2MDNN(
nodes=mol_net_architecture,
elements=[1, 6, 7, 8],
device=self.device,
)
self.atom_net = modules.TorchElementSpecificA2MDNN(
nodes=atom_net_architecture,
elements=[1, 6, 7, 8],
device=self.device,
distribution=coefficients_distribution['isotropic'])
self.bond_net = modules.TorchPairSpecificA2MDNN(
nodes=bond_net_architecture,
elements=[1, 6, 7, 8],
device=self.device,
distribution=coefficients_distribution['anisotropic'])
if feature_extraction_layer is None:
self.feature_extractor = modules.SymFeats()
else:
self.feature_extractor = modules.TorchaniFeats(
net=feature_extraction_net,
feats_layer=feature_extraction_layer,
device=self.device)
for param in self.feature_extractor.parameters():
param.requires_grad = False
self.normalizer = modules.QPChargeNormalization(device=self.device)
self.density_model = A2MDt(params=A2MD_MODEL, device=self.device)