from typing import Optional
from torch import Tensor, nn
from torch_geometric.nn.dense.linear import Linear
from torch_geometric.typing import Adj, OptTensor
from greatx.functional import spmm
from greatx.nn.layers.gcn_conv import make_gcn_norm, make_self_loops
[docs]class SSGConv(nn.Module):
r"""The simple spectral graph convolutional operator from
the `"Simple Spectral Graph Convolution"
<https://openreview.net/forum?id=CYO5T-YjWZV>`_ paper (ICLR'21)
Parameters
----------
in_channels : int
dimensions of int samples
out_channels : int
dimensions of output samples
K : int
the number of propagation steps, by default 5
alpha : float
Teleport probability :math:`\alpha`, by default 0.1
cached : bool, optional
whether the layer will cache
the K-step aggregation on first execution, and will use the
cached version for further executions, by default False
add_self_loops : bool, optional
whether to add self-loops to the input graph, by default True
normalize : bool, optional
whether to compute symmetric normalization
coefficients on the fly, by default True
bias : bool, optional
whether to use bias in the layers, by default True
Note
----
Different from that in :class:`torch_geometric`,
for the input :obj:`edge_index`, our implementation supports
:obj:`torch.FloatTensor`, :obj:`torch.LongTensor`
and obj:`torch_sparse.SparseTensor`.
See also
--------
:class:`greatx.nn.models.supervised.SSGC`
"""
_cached_x: Optional[Tensor]
def __init__(self, in_channels: int, out_channels: int, K: int = 5,
alpha: float = 0.1, cached: bool = False,
add_self_loops: bool = True, normalize: bool = True,
bias: bool = True):
super().__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.K = K
self.alpha = alpha
self.cached = cached
self.add_self_loops = add_self_loops
self.normalize = normalize
self._cached_x = None
self.lin = Linear(in_channels, out_channels, bias=bias,
weight_initializer='glorot')
self.reset_parameters()
[docs] def reset_parameters(self):
self.lin.reset_parameters()
self.cache_clear()
[docs] def cache_clear(self):
"""Clear cached inputs or intermediate results."""
self._cached_x = None
return self
[docs] def forward(self, x: Tensor, edge_index: Adj,
edge_weight: OptTensor = None) -> Tensor:
""""""
cache = self._cached_x
if cache is None:
if self.add_self_loops:
edge_index, edge_weight = make_self_loops(
edge_index, edge_weight, num_nodes=x.size(0))
if self.normalize:
edge_index, edge_weight = make_gcn_norm(
edge_index, edge_weight, num_nodes=x.size(0),
dtype=x.dtype, add_self_loops=False)
x_out = x * self.alpha
for k in range(self.K):
x = spmm(x, edge_index, edge_weight)
x_out = x_out + (1 - self.alpha) / self.K * x
if self.cached:
self._cached_x = x_out
else:
x_out = cache.detach()
return self.lin(x_out)
def __repr__(self) -> str:
return (f'{self.__class__.__name__}({self.in_channels}, '
f'{self.out_channels}, K={self.K}, alpha={self.alpha})')