def __init__(self,
graph,
walk_len,
window_size,
node_count,
hidden_dim,
neg_num,
batch_size,
s2h=False,
ps_hosts=None,
temperature=1.0,
node_type='item',
edge_type='relation'):
super(DeepWalk, self).__init__(graph,
batch_size)
self.walk_len = walk_len
self.window_size = window_size
self.node_count = node_count
self.hidden_dim = hidden_dim
self.neg_num = neg_num
self.s2h = s2h
self.ps_hosts=ps_hosts
self.temperature=temperature
self.node_type = node_type
self.edge_type = edge_type
# construct EgoSpecs.
self.ego_spec = gl.EgoSpec(gl.FeatureSpec(0, 0))
# encoders.
self.encoders = self._encoders()
def __init__(self,
graph,
node_count,
hidden_dim,
neg_num,
batch_size,
s2h=False,
ps_hosts=None,
proximity='first_order',
node_type='item',
edge_type='relation'):
super(LINE, self).__init__(graph, batch_size)
self.node_count = node_count
self.hidden_dim = hidden_dim
self.neg_num = neg_num
self.s2h = s2h
self.proximity = proximity
self.ps_hosts = ps_hosts
self.node_type = node_type
self.edge_type = edge_type
# construct EgoSpecs.
self.ego_spec = gl.EgoSpec(gl.FeatureSpec(0, 0))
# encoders.
self.encoders = self._encoders()
def __init__(self,
graph,
neg_num,
batch_size,
margin,
entity_num,
relation_num,
hidden_dim,
s2h=False,
ps_hosts=None):
super(TransE, self).__init__(graph, batch_size)
self.margin = margin
self.entity_num = entity_num
self.relation_num = relation_num
self.hidden_dim = hidden_dim
self.s2h = s2h
self.neg_num = neg_num
self.ps_hosts = ps_hosts
self.node_ego_spec = gl.EgoSpec(gl.FeatureSpec(1, 0))
# continuous features as edge ids.
self.edge_ego_spec = gl.EgoSpec(gl.FeatureSpec(1, 0))
self.encoders = self._encoders()
def __init__(self,
graph,
output_dim,
features_num,
batch_size,
val_batch_size=None,
test_batch_size=None,
categorical_attrs_desc='',
hidden_dim=16,
in_drop_rate=0,
hops_num=2,
neighs_num=None,
agg_type='gcn',
full_graph_mode=False,
unsupervised=False,
neg_num=10,
node_type='item',
edge_type='relation'):
super(GraphSage, self).__init__(graph,
batch_size,
full_graph_mode=full_graph_mode)
self.features_num = features_num
self.hidden_dim = hidden_dim
self.in_drop_rate = in_drop_rate
self.output_dim = output_dim
self.neighs_num = neighs_num
self.val_batch_size = val_batch_size if val_batch_size else batch_size
self.test_batch_size = test_batch_size if test_batch_size else batch_size
self.agg_type = agg_type
self.hops_num = hops_num
self.unsupervised = unsupervised
self.neg_num = neg_num
self.node_type = node_type
self.edge_type = edge_type
# construct EgoSpecs.
categorical_attrs_num = len(categorical_attrs_desc)
continuous_attrs_num = features_num - categorical_attrs_num
src_spec = gl.FeatureSpec(continuous_attrs_num,
categorical_attrs_num,
labeled=not self.unsupervised)
hop_spec = gl.HopSpec(
gl.FeatureSpec(continuous_attrs_num, categorical_attrs_num))
self.src_ego_spec = gl.EgoSpec(src_spec,
hops_spec=[hop_spec] * self.hops_num)
# encoders.
self.encoders = self._encoders()
def __init__(self,
graph,
output_dim,
features_num,
num_heads,
batch_size,
val_batch_size=None,
test_batch_size=None,
categorical_attrs_desc='',
hidden_dim=16,
in_drop_rate=0,
attn_drop_rate=0,
hops_num=2,
neighs_num=None,
full_graph_mode=False,
node_type='item',
edge_type='relation'):
super(GAT, self).__init__(graph,
batch_size,
full_graph_mode=full_graph_mode)
self.features_num = features_num
self.categorical_attrs_desc = categorical_attrs_desc
self.hops_num = hops_num
self.hidden_dim = hidden_dim
self.num_heads = num_heads
self.in_drop_rate = in_drop_rate
self.attn_drop_rate = attn_drop_rate
self.output_dim = output_dim
self.neighs_num = neighs_num
self.val_batch_size = val_batch_size if val_batch_size else batch_size
self.test_batch_size = test_batch_size if test_batch_size else batch_size
self.node_type = node_type
self.edge_type = edge_type
# construct EgoSpecs.
categorical_attrs_num = len(categorical_attrs_desc)
continuous_attrs_num = features_num - categorical_attrs_num
src_spec = gl.FeatureSpec(continuous_attrs_num,
categorical_attrs_num,
labeled=True)
hop_spec = gl.HopSpec(gl.FeatureSpec(continuous_attrs_num,
categorical_attrs_num),
sparse=(self.neighs_num is None))
self.src_ego_spec = gl.EgoSpec(src_spec,
hops_spec=[hop_spec] * self.hops_num)
# encoders.
self.encoders = self._encoders()
def __init__(self,
graph,
batch_size,
hidden_dim,
output_dim,
hops_num,
u_neighs_num,
i_neighs_num,
u_features_num=0,
u_categorical_attrs_desc='',
i_features_num=0,
i_categorical_attrs_desc='',
neg_num=10,
use_input_bn=True,
act=tf.nn.leaky_relu,
agg_type='gcn',
need_dense=True,
in_drop_rate=0.5,
ps_hosts=None):
super(BipartiteGraphSage, self).__init__(graph, batch_size)
self.hops_num = hops_num
self.neg_num = neg_num
self.hidden_dim = hidden_dim
self.output_dim = output_dim
self.u_neighs_num = u_neighs_num
self.i_neighs_num = i_neighs_num
self.u_features_num = u_features_num
self.u_categorical_attrs_desc = u_categorical_attrs_desc
self.i_features_num = i_features_num
self.i_categorical_attrs_desc = i_categorical_attrs_desc
self.use_input_bn = use_input_bn
self.act = act
self.agg_type = agg_type
self.need_dense = need_dense
self.in_drop_rate = in_drop_rate
self.ps_hosts = ps_hosts
# construct EgoSpecs.
u_categorical_attrs_num = len(u_categorical_attrs_desc)
u_continuous_attrs_num = u_features_num - u_categorical_attrs_num
i_categorical_attrs_num = len(i_categorical_attrs_desc)
i_continuous_attrs_num = i_features_num - i_categorical_attrs_num
u_spec = gl.FeatureSpec(u_continuous_attrs_num,
u_categorical_attrs_num)
i_spec = gl.FeatureSpec(i_continuous_attrs_num,
i_categorical_attrs_num)
u_hop_spec = gl.HopSpec(i_spec)
i_hop_spec = gl.HopSpec(u_spec)
u_hops_spec = [] # u-i-u-i-u...
i_hops_spec = [] # i-u-i-u-i...
for i in range(self.hops_num):
if (i % 2) == 0:
u_hops_spec.append(u_hop_spec)
i_hops_spec.append(i_hop_spec)
else:
u_hops_spec.append(i_hop_spec)
i_hops_spec.append(u_hop_spec)
self.u_ego_spec = gl.EgoSpec(u_spec, hops_spec=u_hops_spec)
self.i_ego_spec = gl.EgoSpec(i_spec, hops_spec=i_hops_spec)
# encoders.
self.encoders = self._encoders()