def __next__(self):
"""Get next batch
Returns
-------
DGLGraph
Sampled positive graph
ChunkNegEdgeSubgraph
Negative graph wrapper
"""
while True:
pos_g, neg_g = next(self.sampler_iter)
if self.filter_false_neg:
neg_positive = neg_g.edata['false_neg']
neg_g = create_neg_subgraph(pos_g, neg_g, self.neg_chunk_size,
self.neg_sample_size, 'chunk'
in self.mode, self.neg_head,
self.g.number_of_nodes())
if neg_g is not None:
break
pos_g.ndata['id'] = pos_g.parent_nid
neg_g.ndata['id'] = neg_g.parent_nid
pos_g.edata['id'] = pos_g._parent.edata['tid'][pos_g.parent_eid]
if self.filter_false_neg:
neg_g.edata['bias'] = F.astype(-neg_positive, F.float32)
return pos_g, neg_g
def __next__(self):
while True:
pos_g, neg_g = next(self.sampler_iter)
neg_positive = neg_g.edata['false_neg']
neg_g = create_neg_subgraph(pos_g, neg_g, 'PBG' in self.mode,
self.neg_head, self.g.number_of_nodes())
if neg_g is not None:
break
pos_g.copy_from_parent()
neg_g.copy_from_parent()
neg_g.edata['bias'] = F.astype(-neg_positive, F.float32)
return pos_g, neg_g
def edge_sparsemax(graph: dgl.DGLGraph, logits, eids=ALL, norm_by="dst"):
r"""
Description
-----------
Compute edge sparsemax. For a node :math:`i`, edge sparsemax is an operation that computes
.. math::
a_{ij} = \text{ReLU}(z_{ij} - \tau(\z_{i,:}))
where :math:`z_{ij}` is a signal of edge :math:`j\rightarrow i`, also
called logits in the context of sparsemax. :math:`\tau` is a function
that can be found at the `From Softmax to Sparsemax <https://arxiv.org/pdf/1602.02068.pdf>`
paper.
NOTE: currently only homogeneous graphs are supported.
Parameters
----------
graph : DGLGraph
The graph to perform edge sparsemax on.
logits : torch.Tensor
The input edge feature.
eids : torch.Tensor or ALL, optional
A tensor of edge index on which to apply edge sparsemax. If ALL, apply edge
sparsemax on all edges in the graph. Default: ALL.
norm_by : str, could be 'src' or 'dst'
Normalized by source nodes of destination nodes. Default: `dst`.
Returns
-------
Tensor
Sparsemax value.
"""
# we get edge index tensors here since it is
# hard to get edge index with HeteroGraphIndex
# object without other information like edge_type.
row, col = graph.all_edges(order="eid")
assert norm_by in ["dst", "src"]
end_n_ids = col if norm_by == "dst" else row
if not is_all(eids):
eids = astype(eids, graph.idtype)
end_n_ids = end_n_ids[eids]
return EdgeSparsemaxFunction.apply(graph._graph, logits, eids, end_n_ids,
norm_by)