def __init__(self, out_dim, hidden_dim, n_layers, processing_steps=3):
super(MPNNReadout, self).__init__()
with self.init_scope():
self.set2set = Set2Set(in_channels=hidden_dim, n_layers=n_layers)
self.linear1 = links.Linear(None, hidden_dim)
self.linear2 = links.Linear(None, out_dim)
self.out_dim = out_dim
self.hidden_dim = hidden_dim
self.n_layers = n_layers
self.processing_steps = processing_steps
def __init__(self, node_features = None, edge_features = None, target_features = 1, hidden_features = 73, num_layers = 3, s2s_processing_steps = 12, type="regression", dropout=0.5, **kwargs):
super(MPNN_ENN_K_Set2Set, self).__init__()
self.input = nn.Linear(in_features=node_features,out_features=hidden_features)
self.ee = EdgeEncoderMLP( edge_features, hidden_features )
self.mpnn = MPNN_enn(edge_features, hidden_features)
self.mpnn.set_T(num_layers)
self.s2s = Set2Set(hidden_features, s2s_processing_steps, num_layers=1)
self.output = nn.Linear(in_features=hidden_features,out_features=target_features)
self.type = type
self.output_function = get_output_function(type,target_features)
def __init__(self, node_features = None, edge_features = None, target_features = 1, hidden_features = 73, num_layers = 3, s2s_processing_steps = 12, type="regression", dropout=0.5, **kwargs):
super(EdgeRES1_K_Set2Set, self).__init__()
self.mlpin = TransitionMLP( node_features, hidden_features )
self.gcmid = RESKnorm( hidden_features, hidden_features, hidden_features, nlayers = num_layers, residue_layers=1 )
self.mlpout = TransitionMLP( hidden_features, target_features )
self.ee = EdgeEncoderMLP( edge_features, hidden_features )
self.s2s = Set2Set(hidden_features, s2s_processing_steps, num_layers=1)
self.type = type
self.output_function = get_output_function(type,target_features)
def __init__(self, node_features = None, edge_features = None, target_features = 1, hidden_features = 73, num_layers = 3, s2s_processing_steps = 12, type="regression", dropout=0.5, **kwargs):
super(EdgeGCN_K_Set2Set, self).__init__()
self.mlpin = TransitionMLP( node_features, hidden_features )
self.gcmid = nn.ModuleList(
[ EdgeGraphConvolution( hidden_features, hidden_features )
for _ in range(num_layers) ] )
self.mlpout = TransitionMLP( hidden_features, target_features )
self.dropout = dropout
self.ee = EdgeEncoderMLP( edge_features, hidden_features )
self.s2s = Set2Set(hidden_features, s2s_processing_steps, num_layers=1)
self.type = type
self.output_function = get_output_function(type,target_features)
def __init__(self,
input_dim,
hidden_dim,
embedding_dim,
label_dim,
num_layers,
pred_hidden_dims=[],
concat=True,
bn=True,
dropout=0.0,
args=None):
super(GcnSet2SetEncoder, self).__init__(input_dim,
hidden_dim,
embedding_dim,
label_dim,
num_layers,
pred_hidden_dims,
concat,
bn,
dropout,
args=args)
self.s2s = Set2Set(self.pred_input_dim, self.pred_input_dim * 2)
