def __init__(self,
args: TrainArgs,
atom_fdim: int = None,
bond_fdim: int = None):
"""
:param args: A :class:`~chemprop.args.TrainArgs` object containing model arguments.
:param atom_fdim: Atom feature vector dimension.
:param bond_fdim: Bond feature vector dimension.
"""
super(MPN, self).__init__()
self.reaction = args.reaction
self.reaction_solvent = args.reaction_solvent
self.atom_fdim = atom_fdim or get_atom_fdim(
overwrite_default_atom=args.overwrite_default_atom_features,
is_reaction=(self.reaction or self.reaction_solvent))
self.bond_fdim = bond_fdim or get_bond_fdim(
overwrite_default_atom=args.overwrite_default_atom_features,
overwrite_default_bond=args.overwrite_default_bond_features,
atom_messages=args.atom_messages,
is_reaction=(self.reaction or self.reaction_solvent))
self.features_only = args.features_only
self.use_input_features = args.use_input_features
self.device = args.device
self.atom_descriptors = args.atom_descriptors
self.overwrite_default_atom_features = args.overwrite_default_atom_features
self.overwrite_default_bond_features = args.overwrite_default_bond_features
if self.features_only:
return
if not self.reaction_solvent:
if args.mpn_shared:
self.encoder = nn.ModuleList(
[MPNEncoder(args, self.atom_fdim, self.bond_fdim)] *
args.number_of_molecules)
else:
self.encoder = nn.ModuleList([
MPNEncoder(args, self.atom_fdim, self.bond_fdim)
for _ in range(args.number_of_molecules)
])
else:
self.encoder = MPNEncoder(args, self.atom_fdim, self.bond_fdim)
# Set separate atom_fdim and bond_fdim for solvent molecules
self.atom_fdim_solvent = get_atom_fdim(
overwrite_default_atom=args.overwrite_default_atom_features,
is_reaction=False)
self.bond_fdim_solvent = get_bond_fdim(
overwrite_default_atom=args.overwrite_default_atom_features,
overwrite_default_bond=args.overwrite_default_bond_features,
atom_messages=args.atom_messages,
is_reaction=False)
self.encoder_solvent = MPNEncoder(args, self.atom_fdim_solvent,
self.bond_fdim_solvent,
args.hidden_size_solvent,
args.bias_solvent,
args.depth_solvent)
def __init__(self,
args: TrainArgs,
atom_fdim: int = None,
bond_fdim: int = None):
"""
:param args: A :class:`~chemprop.args.TrainArgs` object containing model arguments.
:param atom_fdim: Atom feature vector dimension.
:param bond_fdim: Bond feature vector dimension.
"""
super(MPN, self).__init__()
self.atom_fdim = atom_fdim or get_atom_fdim()
self.bond_fdim = bond_fdim or get_bond_fdim(atom_messages=args.atom_messages)
self.features_only = args.features_only
self.use_input_features = args.use_input_features
self.device = args.device
self.atom_descriptors = args.atom_descriptors
if self.features_only:
return
if args.mpn_shared:
self.encoder = nn.ModuleList([MPNEncoder(args, self.atom_fdim, self.bond_fdim)] * args.number_of_molecules)
else:
self.encoder = nn.ModuleList([MPNEncoder(args, self.atom_fdim, self.bond_fdim)
for _ in range(args.number_of_molecules)])
def __init__(self,
args: Namespace,
atom_fdim: int = None,
bond_fdim: int = None,
graph_input: bool = False):
super(MPN, self).__init__()
self.args = args
self.atom_fdim = atom_fdim or get_atom_fdim(args)
self.bond_fdim = bond_fdim or get_bond_fdim(args) + \
(not args.atom_messages) * self.atom_fdim # * 2
self.graph_input = graph_input
self.encoder = MPNEncoder(self.args, self.atom_fdim, self.bond_fdim)
def __init__(self,
args: TrainArgs,
atom_fdim: int = None,
bond_fdim: int = None):
"""
Initializes the MPN.
:param args: Arguments.
:param atom_fdim: Atom features dimension.
:param bond_fdim: Bond features dimension.
"""
super(MPN, self).__init__()
self.atom_fdim = atom_fdim or get_atom_fdim()
self.bond_fdim = bond_fdim or get_bond_fdim(atom_messages=args.atom_messages)
print(get_atom_fdim())
print(get_bond_fdim(atom_messages=args.atom_messages))
# self.atom_fdim = 127
# self.bond_fdim = 12
self.encoder = MPNEncoder(args, self.atom_fdim, self.bond_fdim)
def __init__(self,
args: TrainArgs,
atom_fdim: int = None,
bond_fdim: int = None):
"""
:param args: A :class:`~chemprop.args.TrainArgs` object containing model arguments.
:param atom_fdim: Atom feature vector dimension.
:param bond_fdim: Bond feature vector dimension.
"""
super(MPN, self).__init__()
self.atom_fdim = atom_fdim or get_atom_fdim()
self.bond_fdim = bond_fdim or get_bond_fdim(
atom_messages=args.atom_messages)
self.encoder = MPNEncoder(args, self.atom_fdim, self.bond_fdim)
def __init__(self,
args: Namespace,
atom_fdim: int = None,
bond_fdim: int = None,
graph_input: bool = False,
params: Dict[str, nn.Parameter] = None):
super(MPN, self).__init__()
self.args = args
self.atom_fdim = atom_fdim or get_atom_fdim(args)
self.bond_fdim = bond_fdim or get_bond_fdim(
args) + (not args.atom_messages) * self.atom_fdim
self.graph_input = graph_input
self.encoder = MPNEncoder(self.args,
self.atom_fdim,
self.bond_fdim,
params=params)
def __init__(self,
args: Namespace,
atom_fdim: int = None,
bond_fdim: int = None,
graph_input: bool = False):
"""
Initializes the MPN.
:param args: Arguments.
:param atom_fdim: Atom features dimension.
:param bond_fdim: Bond features dimension.
:param graph_input: If true, expects BatchMolGraph as input. Otherwise expects a list of smiles strings as input.
"""
super(MPN, self).__init__()
self.args = args
self.atom_fdim = atom_fdim or get_atom_fdim(args)
self.bond_fdim = bond_fdim or get_bond_fdim(args) + (not args.atom_messages) * self.atom_fdim
self.graph_input = graph_input
self.encoder = MPNEncoder(self.args, self.atom_fdim, self.bond_fdim)
def __init__(self, args: Namespace, smiles: List[str]):
self.substructure_sizes = args.bert_substructure_sizes
self.vocab_func = partial(atom_vocab,
vocab_func=args.bert_vocab_func,
substructure_sizes=self.substructure_sizes,
args=args)
if args.bert_vocab_func == 'feature_vector':
self.unk = None
self.output_size = get_atom_fdim(args, is_output=True)
return # don't need a real vocab list here
self.unk = 'unk'
self.smiles = smiles
self.vocab = get_vocab(args, self.vocab_func, self.smiles)
self.vocab.add(self.unk)
self.vocab_size = len(self.vocab)
self.vocab_mapping = {
word: i
for i, word in enumerate(sorted(self.vocab))
}
self.output_size = self.vocab_size
