def __init__(self):
super(Model, self).__init__()
#lift features of the nodes
self.lifting_layer = nn.Embedding(hyperparams["num_features"],
hyperparams["hsz"])
#latent representations of the nodes
self.sageConv1 = SAGEConv(in_feats = hyperparams["hsz"], \
out_feats = hyperparams["hsz"], \
aggregator_type = 'lstm')
self.sageConv2 = SAGEConv(in_feats = hyperparams["hsz"], \
out_feats = hyperparams["hsz"], \
aggregator_type = 'lstm')
self.sageConv3 = SAGEConv(in_feats = hyperparams["hsz"], \
out_feats = hyperparams["hsz"], \
aggregator_type = 'lstm')
self.GAT_conv1 = GATConv(in_feats = hyperparams["hsz"], \
out_feats = hyperparams["hsz"], \
num_heads = hyperparams["num_heads"])
#readout layer (also task specific layer during pretraining)
self.output_layer = nn.Linear(hyperparams["hsz"], 3)
def __init__(self, in_feats, n_hidden, n_classes, n_layers, activation,
dropout, aggregator_type):
super(GraphSAGE_model, self).__init__()
self.layers = nn.ModuleList()
# input layer
self.layers.append(
SAGEConv(in_feats,
n_hidden,
aggregator_type,
feat_drop=0.,
activation=activation))
# hidden layers
for i in range(n_layers - 1):
self.layers.append(
SAGEConv(n_hidden,
n_hidden,
aggregator_type,
feat_drop=dropout,
activation=activation))
# output layer
self.layers.append(
SAGEConv(n_hidden,
n_classes,
aggregator_type,
feat_drop=dropout,
activation=None))
def __init__(self):
super().__init__()
self.layer1 = SAGEConv(4, 32, aggregator_type='mean')
self.layer2 = SAGEConv(32, 32, aggregator_type='mean')
self.layer3 = SAGEConv(32, 32, aggregator_type='mean')
self.layer4 = SAGEConv(32, 32, aggregator_type='mean')
self.layer5 = SAGEConv(32, 3, aggregator_type='mean')
def __init__(self, in_feats, h_feats, aggr, num_layers):
super(GraphSAGE, self).__init__()
self.conv1 = SAGEConv(in_feats, h_feats, aggr)
# self.convs = torch.nn.ModuleList()
# for i in range(num_layers - 2):
# self.convs.append(SAGEConv(h_feats, h_feats, aggr))
self.conv2 = SAGEConv(h_feats, h_feats, aggr)
self.h_feats = h_feats
def __init__(self, in_features, hidden_features, out_features, activation,
dropout, aggregator):
super().__init__()
self.conv1 = SAGEConv(in_features, hidden_features, aggregator)
self.conv2 = SAGEConv(hidden_features, hidden_features, aggregator)
self.conv3 = SAGEConv(hidden_features, out_features, aggregator)
self.dropout = nn.Dropout(p=dropout)
self.bn1 = torch.nn.BatchNorm1d(hidden_features)
self.bn2 = torch.nn.BatchNorm1d(hidden_features)
self.activation = F.elu_ if activation == 'Elu' else F.relu
def __init__(self,
g,
in_feats,
n_hidden,
n_classes,
aggr,
activation=F.relu,
dropout=0.):
super(GraphSAGE, self).__init__()
self.layers = nn.ModuleList()
self.g = g
self.layers.append(SAGEConv(in_feats, n_hidden, aggr, activation=activation, bias=False))
self.layers.append(SAGEConv(n_hidden, n_classes, aggr, feat_drop=dropout, activation=None, bias=False))
def __init__(self,
in_features,
hidden_features,
out_features,
aggregator,
activation,
dropout=0.5,
is_out_layer=True):
super().__init__()
# Fixme: Deal zero degree
self.conv1 = SAGEConv(in_features, hidden_features, aggregator)
self.conv2 = SAGEConv(hidden_features, out_features, aggregator)
self.dropout = nn.Dropout(p=dropout)
self.is_out_layer = is_out_layer
self.activation = F.elu_ if activation == 'Elu' else F.relu
def __init__(self, in_feats, n_hidden, n_classes, n_layers, activation,
dropout, aggregator_type):
super(GraphSAGE, self).__init__()
self.layers = nn.ModuleList()
self.dropout = nn.Dropout(dropout)
self.activation = activation
# input layer
self.layers.append(SAGEConv(in_feats, n_hidden, aggregator_type))
# hidden layers
for i in range(n_layers - 1):
self.layers.append(SAGEConv(n_hidden, n_hidden, aggregator_type))
# output layer
self.layers.append(SAGEConv(n_hidden, n_classes,
aggregator_type)) # activation None
def __init__(self,
in_feats,
hidden_feats=None,
activation=None,
dropout=None,
aggregator_type=None):
super(GraphSAGE, self).__init__()
if hidden_feats is None:
hidden_feats = [64, 64]
n_layers = len(hidden_feats)
if activation is None:
activation = [F.relu for _ in range(n_layers)]
if dropout is None:
dropout = [0. for _ in range(n_layers)]
if aggregator_type is None:
aggregator_type = ['mean' for _ in range(n_layers)]
lengths = [len(hidden_feats), len(activation), len(dropout), len(aggregator_type)]
assert len(set(lengths)) == 1, 'Expect the lengths of hidden_feats, activation, ' \
'dropout and aggregator_type to be the same, ' \
'got {}'.format(lengths)
self.hidden_feats = hidden_feats
self.gnn_layers = nn.ModuleList()
for i in range(n_layers):
self.gnn_layers.append(SAGEConv(in_feats, hidden_feats[i], aggregator_type[i],
dropout[i], activation[i]))
in_feats = hidden_feats[i]
def __init__(self, in_feats, n_hidden, n_classes, n_layers, activation,
dropout, aggregator):
super().__init__()
self.n_layers = n_layers
self.n_hidden = n_hidden
self.n_classes = n_classes
self.layers = nn.ModuleList()
self.bns = nn.ModuleList()
self.layers.append(SAGEConv(in_feats, n_hidden, aggregator))
self.bns.append(torch.nn.BatchNorm1d(n_hidden))
for i in range(1, n_layers - 1):
self.layers.append(SAGEConv(n_hidden, n_hidden, aggregator))
self.bns.append(torch.nn.BatchNorm1d(n_hidden))
self.layers.append(SAGEConv(n_hidden, n_classes, aggregator))
self.dropout = nn.Dropout(dropout)
self.activation = activation
def __init__(self, in_feats, out_feats, activation, dropout,
aggregator_type, batch_norm, residual=False,
bias=True, dgl_builtin=False):
super().__init__()
self.in_channels = in_feats
self.out_channels = out_feats
self.aggregator_type = aggregator_type
self.batch_norm = batch_norm
self.residual = residual
self.dgl_builtin = dgl_builtin
if in_feats != out_feats:
self.residual = False
self.dropout = nn.Dropout(p=dropout)
if dgl_builtin == False:
self.nodeapply = NodeApply(in_feats, out_feats, activation, dropout,
bias=bias)
if aggregator_type == "maxpool":
self.aggregator = MaxPoolAggregator(in_feats, in_feats,
activation, bias)
elif aggregator_type == "lstm":
self.aggregator = LSTMAggregator(in_feats, in_feats)
else:
self.aggregator = MeanAggregator()
else:
self.sageconv = SAGEConv(in_feats, out_feats, aggregator_type,
dropout, activation=activation)
if self.batch_norm:
self.batchnorm_h = nn.BatchNorm1d(out_feats)
def __init__(self, input_dim, target_dim):
super(NodeClassifier, self).__init__()
hidden_dim = 32
self.graph_conv1 = SAGEConv(input_dim,
hidden_dim,
aggregator_type='pool')
self.graph_conv2 = SAGEConv(hidden_dim,
hidden_dim,
aggregator_type='pool')
self.graph_conv3 = SAGEConv(hidden_dim,
target_dim,
aggregator_type='pool')
self.relu = nn.ReLU()
self.dropout = nn.Dropout(p=0.2)
self.sigmoid = nn.Sigmoid()
def __init__(self, in_feats, hidden_size, hidden_size1, hidden_size2, hidden_size3, hidden_size4, num_classes):
super(GCN, self).__init__()
self.conv1 = SAGEConv(in_feats, hidden_size, aggregator_type='pool', activation=torch.tanh)
self.conv2 = ChebConv(hidden_size, hidden_size1, 4)
self.conv3 = TAGConv(hidden_size1, hidden_size2, activation=F.leaky_relu, k=3)
self.conv4 = SAGEConv(hidden_size2, hidden_size3, aggregator_type='pool', activation=torch.tanh)
self.conv5 = ChebConv(hidden_size3, hidden_size4, 4)
self.conv6 = TAGConv(hidden_size4, num_classes, activation=F.leaky_relu, k=3)
x = 150
self.encoder = nn.Sequential(
nn.Conv2d(1, x, (3, 3)),
nn.LeakyReLU(),
nn.Dropout2d(),
nn.Conv2d(x, 2*x, (3, 2)),
nn.LeakyReLU(),
nn.Dropout2d(),
nn.Conv2d(2*x, 1, (3, 2)),
)
def __init__(self, in_feats, hidden_feats, out_feats, num_layers, dropout):
super(GraphSAGE, self).__init__()
self.layers = nn.ModuleList()
self.bns = nn.ModuleList()
# input layer
self.layers.append(SAGEConv(in_feats, hidden_feats, 'mean',
bias=False))
self.bns.append(nn.BatchNorm1d(hidden_feats))
# hidden layers
for _ in range(num_layers - 2):
self.layers.append(
SAGEConv(hidden_feats, hidden_feats, 'mean', bias=False))
self.bns.append(nn.BatchNorm1d(hidden_feats))
# output layer
self.layers.append(
SAGEConv(hidden_feats, out_feats, 'mean', bias=False))
self.dropout = nn.Dropout(p=dropout)
def __init__(self, num_features, hidden_channels, num_layers, num_classes):
super(SAGE, self).__init__()
self.num_layers = num_layers
self.convs = torch.nn.ModuleList()
for i in range(num_layers):
inc = outc = hidden_channels
if i == 0:
inc = num_features
if i == num_layers - 1:
outc = num_classes
self.convs.append(SAGEConv(inc, outc, "gcn"))
self.dropout = torch.nn.Dropout()
def __init__(self):
super(Model, self).__init__()
###########shared layers###########
self.lifting_layer = nn.Embedding(hyperparams["num_features"],
hyperparams["hsz"])
self.sageConv1 = SAGEConv(in_feats = hyperparams["hsz"], \
out_feats = hyperparams["hsz"], \
aggregator_type = 'lstm')
self.sageConv2 = SAGEConv(in_feats = hyperparams["hsz"], \
out_feats = hyperparams["hsz"], \
aggregator_type = 'lstm')
self.sageConv3 = SAGEConv(in_feats = hyperparams["hsz"], \
out_feats = hyperparams["hsz"], \
aggregator_type = 'lstm')
self.GAT_conv1 = GATConv(in_feats = hyperparams["hsz"], \
out_feats = hyperparams["hsz"], \
num_heads = hyperparams["num_heads"])
###################################
#######task specific layer(s)#######
self.readout_layer = nn.Linear(hyperparams["hsz"], 1)
def __init__(self, in_channels, out_channels, model):
super(CoNet, self).__init__()
if model == 'AFFN':
self.layer1 = SAGEConv(in_channels, out_channels, 'mean')
self.layer2 = GraphConv(in_channels, out_channels)
self.layer3 = GATConv(in_channels, out_channels, 1)
elif model == 'GCN':
self.layer1 = GraphConv(in_channels, out_channels)
self.layer2 = GraphConv(in_channels, out_channels)
self.layer3 = GraphConv(in_channels, out_channels)
elif model == 'SAGE':
self.layer1 = SAGEConv(in_channels, out_channels, 'mean')
self.layer2 = SAGEConv(in_channels, out_channels, 'mean')
self.layer3 = SAGEConv(in_channels, out_channels, 'mean')
else:
self.layer1 = GATConv(in_channels, out_channels, 1)
self.layer2 = GATConv(in_channels, out_channels, 1)
self.layer3 = GATConv(in_channels, out_channels, 1)
self.w = nn.Parameter(th.tensor([1, 1, 1], dtype=th.float))
self.model = model
def __init__(self, g, in_feats, n_hidden, n_classes, n_layers, activation,
dropout, aggregator_type):
super(GraphSAGE, self).__init__()
self.g = g
with self.name_scope():
self.layers = nn.Sequential()
# input layer
self.layers.add_module(
SAGEConv(in_feats,
n_hidden,
aggregator_type,
feat_drop=dropout,
activation=activation))
# hidden layers
for i in range(n_layers - 1):
self.layers.add_module(
SAGEConv(n_hidden,
n_hidden,
aggregator_type,
feat_drop=dropout,
activation=activation))
# output layer
self.layers.add_module(nn.Linear(n_hidden, n_classes))
def __init__(self,
g,
in_feats,
n_hidden,
n_classes,
n_layers,
activation,
dropout,
aspect_embed_size,
aggregator_type='pool'):
super(GraphSAGE_post, self).__init__()
self.g = g
self.layers = nn.ModuleList()
self.n_layers = n_layers
self.layers.append(SAGEConv(in_feats, n_hidden, aggregator_type, \
activation=activation))
for i in range(n_layers - 2):
self.layers.append(SAGEConv(n_hidden, n_hidden, aggregator_type, \
activation=activation))
self.layers.append(
SAGEConv(n_hidden, aspect_embed_size, aggregator_type))
self.layers.append(
SAGEConv(aspect_embed_size, n_classes, aggregator_type))
self.dropout = nn.Dropout(p=dropout)
def __init__(self, device='cpu', dim=64, model_type="gcn"): #初始化
super().__init__()
self.dim = dim
self.device = device
self.embed = nn.Embedding(500, dim)
if model_type == "gcn":
self.conv = GraphConv(dim, dim)
elif model_type == "gat":
self.conv = GATConv(dim, dim, 1)
elif model_type == "gin":
print("gin")
self.conv = GINConv(apply_func=nn.Linear(dim, dim),
aggregator_type="mean")
elif model_type == "sage":
print("sage")
self.conv = SAGEConv(dim, dim, aggregator_type="gcn")
self.func = nn.Linear(dim, 2)
self.act = nn.ReLU()
def __init__(self,
in_feats,
out_feats,
conv_type,
activation=None,
residual=True,
batchnorm=True,
dropout=0.,
num_heads=1,
negative_slope=0.2):
super(ConvLayer, self).__init__()
self.activation = activation
self.conv_type = conv_type
if conv_type == 'gcn':
self.graph_conv = GraphConv(in_feats=in_feats,
out_feats=out_feats,
norm='both',
activation=activation)
elif conv_type == 'sage':
self.graph_conv = SAGEConv(in_feats=in_feats,
out_feats=out_feats,
aggregator_type='mean',
norm=None,
activation=activation)
elif conv_type == 'gat':
assert out_feats % num_heads == 0
self.graph_conv = GATConv(in_feats=in_feats,
out_feats=out_feats // num_heads,
num_heads=num_heads,
feat_drop=dropout,
attn_drop=dropout,
negative_slope=negative_slope,
activation=activation)
self.dropout = nn.Dropout(dropout)
self.residual = residual
if residual:
self.res_connection = nn.Linear(in_feats, out_feats)
self.bn = batchnorm
if batchnorm:
self.bn_layer = nn.BatchNorm1d(out_feats)
def track_time(graph_name, feat_dim, aggr_type):
device = utils.get_bench_device()
graph = utils.get_graph(graph_name).to(device)
feat = torch.randn((graph.num_nodes(), feat_dim), device=device)
model = SAGEConv(feat_dim,
feat_dim,
aggr_type,
activation=F.relu,
bias=False).to(device)
# dry run
for i in range(3):
model(graph, feat)
# timing
with utils.Timer() as t:
for i in range(50):
model(graph, feat)
return t.elapsed_secs / 50
def track_time(feat_dim, num_relations):
device = utils.get_bench_device()
dd = {}
nn_dict = {}
candidate_edges = [
dgl.data.CoraGraphDataset(verbose=False)[0].edges(),
dgl.data.PubmedGraphDataset(verbose=False)[0].edges(),
dgl.data.CiteseerGraphDataset(verbose=False)[0].edges()
]
for i in range(num_relations):
dd[('n1', 'e_{}'.format(i),
'n2')] = candidate_edges[i % len(candidate_edges)]
nn_dict['e_{}'.format(i)] = SAGEConv(feat_dim,
feat_dim,
'mean',
activation=F.relu)
# dry run
feat_dict = {}
graph = dgl.heterograph(dd)
for i in range(num_relations):
etype = 'e_{}'.format(i)
feat_dict[etype] = torch.randn((graph[etype].num_nodes(), feat_dim),
device=device)
conv = HeteroGraphConv(nn_dict).to(device)
# dry run
for i in range(3):
conv(graph, feat_dict)
# timing
with utils.Timer() as t:
for i in range(50):
conv(graph, feat_dict)
return t.elapsed_secs / 50
def __init__(self, vocablen, embedding_dim):
super(Encode_Graph, self).__init__()
self.embed = nn.Embedding(vocablen, embedding_dim)
self.gcn1 = GraphConv(16, 32)
self.gcn2 = SAGEConv(32, 64, aggregator_type='pool')
self.linear1 = nn.Linear(64, 64)
def __init__(self,
num_layers,
hidden_units,
k=10,
gcn_type='gcn',
node_attributes=None,
edge_weights=None,
node_embedding=None,
use_embedding=False,
num_nodes=None,
dropout=0.5,
max_z=1000):
super(DGCNN, self).__init__()
self.num_layers = num_layers
self.dropout = dropout
self.use_attribute = False if node_attributes is None else True
self.use_embedding = use_embedding
self.use_edge_weight = False if edge_weights is None else True
self.z_embedding = nn.Embedding(max_z, hidden_units)
if node_attributes is not None:
self.node_attributes_lookup = nn.Embedding.from_pretrained(
node_attributes)
self.node_attributes_lookup.weight.requires_grad = False
if edge_weights is not None:
self.edge_weights_lookup = nn.Embedding.from_pretrained(
edge_weights)
self.edge_weights_lookup.weight.requires_grad = False
if node_embedding is not None:
self.node_embedding = nn.Embedding.from_pretrained(node_embedding)
self.node_embedding.weight.requires_grad = False
elif use_embedding:
self.node_embedding = nn.Embedding(num_nodes, hidden_units)
initial_dim = hidden_units
if self.use_attribute:
initial_dim += self.node_attributes_lookup.embedding_dim
if self.use_embedding:
initial_dim += self.node_embedding.embedding_dim
self.layers = nn.ModuleList()
if gcn_type == 'gcn':
self.layers.append(
GraphConv(initial_dim, hidden_units,
allow_zero_in_degree=True))
for _ in range(num_layers - 1):
self.layers.append(
GraphConv(hidden_units,
hidden_units,
allow_zero_in_degree=True))
self.layers.append(
GraphConv(hidden_units, 1, allow_zero_in_degree=True))
elif gcn_type == 'sage':
self.layers.append(
SAGEConv(initial_dim, hidden_units, aggregator_type='gcn'))
for _ in range(num_layers - 1):
self.layers.append(
SAGEConv(hidden_units, hidden_units,
aggregator_type='gcn'))
self.layers.append(SAGEConv(hidden_units, 1,
aggregator_type='gcn'))
else:
raise ValueError('Gcn type error.')
self.pooling = SortPooling(k=k)
conv1d_channels = [16, 32]
total_latent_dim = hidden_units * num_layers + 1
conv1d_kws = [total_latent_dim, 5]
self.conv_1 = nn.Conv1d(1, conv1d_channels[0], conv1d_kws[0],
conv1d_kws[0])
self.maxpool1d = nn.MaxPool1d(2, 2)
self.conv_2 = nn.Conv1d(conv1d_channels[0], conv1d_channels[1],
conv1d_kws[1], 1)
dense_dim = int((k - 2) / 2 + 1)
dense_dim = (dense_dim - conv1d_kws[1] + 1) * conv1d_channels[1]
self.linear_1 = nn.Linear(dense_dim, 128)
self.linear_2 = nn.Linear(128, 1)
def __init__(self, k, feature_dims, emb_dims, output_classes, init_points = 512, input_dims=3,
dropout_prob=0.5, npart=1, id_skip=False, drop_connect_rate=0, res_scale = 1.0,
light = False, bias = False, cluster='xyz', conv='EdgeConv', use_xyz=True, use_se = True, graph_jitter = False):
super(Model, self).__init__()
self.npart = npart
self.graph_jitter = graph_jitter
self.res_scale = res_scale
self.id_skip = id_skip
self.drop_connect_rate = drop_connect_rate
self.nng = KNNGraphE(k) # with random neighbor
self.conv = nn.ModuleList()
self.conv_s1 = nn.ModuleList()
self.conv_s2 = nn.ModuleList()
self.bn = nn.ModuleList()
self.sa = nn.ModuleList()
self.cluster = cluster
self.feature_dims = feature_dims
self.conv_type = conv
self.init_points = init_points
self.k = k
#self.proj_in = nn.Linear(input_dims, input_dims)
self.num_layers = len(feature_dims)
npoint = init_points
for i in range(self.num_layers):
if k==1:
self.conv.append(nn.Linear(feature_dims[i-1] if i > 0 else input_dims,
feature_dims[i] ))
elif conv == 'EdgeConv':
if light:
self.conv.append(EdgeConv_Light(
feature_dims[i - 1] if i > 0 else input_dims,
feature_dims[i],
batch_norm=True))
else:
self.conv.append(EdgeConv(
feature_dims[i - 1] if i > 0 else input_dims,
feature_dims[i],
batch_norm=True))
elif conv == 'GATConv':
self.conv.append(GATConv(
feature_dims[i - 1] if i > 0 else input_dims,
feature_dims[i],
feat_drop=0.2, attn_drop=0.2,
residual=True,
num_heads=1))
elif conv == 'GraphConv':
self.conv.append( GraphConv(
feature_dims[i - 1] if i > 0 else input_dims,
feature_dims[i]))
elif conv == 'SAGEConv':
self.conv.append( SAGEConv(
feature_dims[i - 1] if i > 0 else input_dims,
feature_dims[i],
feat_drop=0.2,
aggregator_type='mean',
norm = nn.BatchNorm1d(feature_dims[i])
) )
elif conv == 'SGConv':
self.conv.append( SGConv(
feature_dims[i - 1] if i > 0 else input_dims,
feature_dims[i]) )
elif conv == 'GatedGCN': # missing etypes
self.conv.append( GatedGCNLayer(
feature_dims[i - 1] if i > 0 else input_dims,
feature_dims[i],
dropout=0.0,
graph_norm=True, batch_norm=True, residual=True)
)
if i>0 and feature_dims[i]>feature_dims[i-1]:
npoint = npoint//2
if id_skip and npoint <= self.init_points//4: # Only work on high level
self.conv_s2.append( nn.Linear(feature_dims[i-1], feature_dims[i] ))
self.sa.append(PointnetSAModule(
npoint=npoint,
radius=0.2,
nsample=64,
mlp=[feature_dims[i], feature_dims[i], feature_dims[i]],
fuse = 'add',
norml = 'bn',
activation = 'relu',
use_se = use_se,
use_xyz = use_xyz,
use_neighbor = False,
light = light
))
#if id_skip:
# self.conv_s1.append( nn.Linear(feature_dims[i], feature_dims[i] ))
self.embs = nn.ModuleList()
self.bn_embs = nn.ModuleList()
self.dropouts = nn.ModuleList()
self.partpool = nn.AdaptiveAvgPool1d(self.npart)
if self.npart == 1:
self.embs.append(nn.Linear(
# * 2 because of concatenation of max- and mean-pooling
feature_dims[-1]*2, emb_dims[0], bias=bias))
self.bn_embs.append(nn.BatchNorm1d(emb_dims[0]))
self.dropouts.append(nn.Dropout(dropout_prob, inplace=True))
self.proj_output = nn.Linear(emb_dims[0], output_classes)
self.proj_output.apply(weights_init_classifier)
else:
self.proj_outputs = nn.ModuleList()
for i in range(0, self.npart):
self.embs.append(nn.Linear(512, 512, bias=bias))
self.bn_embs.append(nn.BatchNorm1d(512))
self.dropouts.append(nn.Dropout(dropout_prob, inplace=True))
self.proj_outputs.append(nn.Linear(512, output_classes))
self.proj_outputs.apply(weights_init_classifier)
# initial
#self.proj_in.apply(weights_init_kaiming)
self.conv.apply(weights_init_kaiming)
self.conv_s1.apply(weights_init_kaiming)
self.conv_s2.apply(weights_init_kaiming)
weights_init_kaiming2 = lambda x:weights_init_kaiming(x,L=self.num_layers)
self.sa.apply(weights_init_kaiming2)
#self.proj.apply(weights_init_kaiming)
self.embs.apply(weights_init_kaiming)
self.bn.apply(weights_init_kaiming)
self.bn_embs.apply(weights_init_kaiming)
self.npart = npart
def __init__(self,
num_layers,
hidden_units,
gcn_type='gcn',
pooling_type='sum',
node_attributes=None,
edge_weights=None,
node_embedding=None,
use_embedding=False,
num_nodes=None,
dropout=0.5,
max_z=1000):
super(GCN, self).__init__()
self.num_layers = num_layers
self.dropout = dropout
self.pooling_type = pooling_type
self.use_attribute = False if node_attributes is None else True
self.use_embedding = use_embedding
self.use_edge_weight = False if edge_weights is None else True
self.z_embedding = nn.Embedding(max_z, hidden_units)
if node_attributes is not None:
self.node_attributes_lookup = nn.Embedding.from_pretrained(
node_attributes)
self.node_attributes_lookup.weight.requires_grad = False
if edge_weights is not None:
self.edge_weights_lookup = nn.Embedding.from_pretrained(
edge_weights)
self.edge_weights_lookup.weight.requires_grad = False
if node_embedding is not None:
self.node_embedding = nn.Embedding.from_pretrained(node_embedding)
self.node_embedding.weight.requires_grad = False
elif use_embedding:
self.node_embedding = nn.Embedding(num_nodes, hidden_units)
initial_dim = hidden_units
if self.use_attribute:
initial_dim += self.node_attributes_lookup.embedding_dim
if self.use_embedding:
initial_dim += self.node_embedding.embedding_dim
self.layers = nn.ModuleList()
if gcn_type == 'gcn':
self.layers.append(
GraphConv(initial_dim, hidden_units,
allow_zero_in_degree=True))
for _ in range(num_layers - 1):
self.layers.append(
GraphConv(hidden_units,
hidden_units,
allow_zero_in_degree=True))
elif gcn_type == 'sage':
self.layers.append(
SAGEConv(initial_dim, hidden_units, aggregator_type='gcn'))
for _ in range(num_layers - 1):
self.layers.append(
SAGEConv(hidden_units, hidden_units,
aggregator_type='gcn'))
else:
raise ValueError('Gcn type error.')
self.linear_1 = nn.Linear(hidden_units, hidden_units)
self.linear_2 = nn.Linear(hidden_units, 1)
if pooling_type != 'sum':
raise ValueError('Pooling type error.')
self.pooling = SumPooling()
