# Source code for torch_geometric.nn.dense.dense_sage_conv

import torch
import torch.nn.functional as F
from torch.nn import Linear

[docs]class DenseSAGEConv(torch.nn.Module):
r"""See :class:torch_geometric.nn.conv.SAGEConv.

.. note::

:class:~torch_geometric.nn.dense.DenseSAGEConv expects to work on
If you want to make use of weighted dense adjacency matrices, please
use :class:torch_geometric.nn.dense.DenseGraphConv instead.

"""
def __init__(self, in_channels, out_channels, normalize=False, bias=True):
super(DenseSAGEConv, self).__init__()

self.in_channels = in_channels
self.out_channels = out_channels
self.normalize = normalize

self.lin_rel = Linear(in_channels, out_channels, bias=False)
self.lin_root = Linear(in_channels, out_channels, bias=bias)

self.reset_parameters()

[docs]    def reset_parameters(self):
self.lin_rel.reset_parameters()
self.lin_root.reset_parameters()

r"""
Args:
x (Tensor): Node feature tensor :math:\mathbf{X} \in \mathbb{R}^{B
\times N \times F}, with batch-size :math:B, (maximum)
number of nodes :math:N for each graph, and feature
dimension :math:F.
\times N \times N}. The adjacency tensor is broadcastable in
the batch dimension, resulting in a shared adjacency matrix for
the complete batch.
:math:\mathbf{M} \in {\{ 0, 1 \}}^{B \times N} indicating
the valid nodes for each graph. (default: :obj:None)
"""
x = x.unsqueeze(0) if x.dim() == 2 else x

out = out / adj.sum(dim=-1, keepdim=True).clamp(min=1)
out = self.lin_rel(out) + self.lin_root(x)

if self.normalize:
out = F.normalize(out, p=2, dim=-1)