# Copyright 2022 Huawei Technologies Co., Ltd
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ============================================================================
"""GINConv Layer"""
import mindspore as ms
from mindspore_gl import Graph
from .. import GNNCell
[文档]class GINConv(GNNCell):
r"""
Graph isomorphic network layer.
From the paper `How Powerful are Graph Neural Networks? <https://arxiv.org/pdf/1810.00826.pdf>`_ .
.. math::
h_i^{(l+1)} = f_\Theta \left((1 + \epsilon) h_i^{l} +
\mathrm{aggregate}\left(\left\{h_j^{l}, j\in\mathcal{N}(i)
\right\}\right)\right)
If weights are provided on each edge, the weighted graph convolution is defined as:
.. math::
h_i^{(l+1)} = f_\Theta \left((1 + \epsilon) h_i^{l} +
\mathrm{aggregate}\left(\left\{e_{ji} h_j^{l}, j\in\mathcal{N}(i)
\right\}\right)\right)
Args:
activation (mindspore.nn.Cell): Activation function.
init_eps (float, optional): Init value of eps. Default: ``0``.
learn_eps (bool, optional): Whether eps is learnable. Default: ``False``.
aggregation_type (str, optional): Type of aggregation, should in ``'sum'``, ``'max'`` and ``'avg'``.
Default: ``'sum'``.
Inputs:
- **x** (Tensor): The input node features. The shape is :math:`(N,*)` where :math:`N` is the number of nodes,
and :math:`*` could be of any shape.
- **edge_weight** (Tensor): The input edge weights. The shape is :math:`(M,*)` where :math:`M` is the number
of nodes, and :math:`*` could be of any shape.
- **g** (Graph): The input graph.
Outputs:
- Tensor, output node features. The shape is :math:`(N, out\_feat\_size)`.
Raises:
TypeError: If `activation` is not a `mindspore.nn.Cell`.
TypeError: If `init_eps` is not a float.
TypeError: If `learn_eps` is not a bool.
SyntaxError: Raised when the `aggregation_type` not in ``'sum'``, ``'max'`` and ``'avg'``.
Supported Platforms:
``Ascend`` ``GPU``
Examples:
>>> import mindspore as ms
>>> from mindspore_gl.nn import GINConv
>>> from mindspore_gl import GraphField
>>> n_nodes = 4
>>> n_edges = 8
>>> feat_size = 16
>>> src_idx = ms.Tensor([0, 0, 0, 1, 1, 1, 2, 3], ms.int32)
>>> dst_idx = ms.Tensor([0, 1, 3, 1, 2, 3, 3, 2], ms.int32)
>>> ones = ms.ops.Ones()
>>> nodes_feat = ones((n_nodes, feat_size), ms.float32)
>>> edges_weight = ones((n_edges, feat_size), ms.float32)
>>> graph_field = GraphField(src_idx, dst_idx, n_nodes, n_edges)
>>> conv = GINConv(activation=None, init_eps=0., learn_eps=False, aggregation_type="sum")
>>> ret = conv(nodes_feat, edges_weight, *graph_field.get_graph())
>>> print(ret.shape)
(4, 16)
"""
def __init__(self,
activation,
init_eps=0.,
learn_eps=False,
aggregation_type="sum"):
super().__init__()
if not isinstance(init_eps, float):
raise ValueError("init_eps must be float")
if not isinstance(learn_eps, bool):
raise ValueError("learn_eps must be bool")
if activation is not None and not isinstance(activation, ms.nn.Cell):
raise TypeError(f"For '{self.cls_name}', the 'activation' must a mindspore.nn.Cell, but got "
f"{type(activation).__name__}.")
self.agg_type = aggregation_type
if aggregation_type not in {"sum", "max", "avg"}:
raise SyntaxError("Aggregation type must be one of sum, max or avg")
if learn_eps:
self.eps = ms.Parameter(ms.Tensor(init_eps, ms.float32))
else:
self.eps = ms.Tensor(init_eps, ms.float32)
self.act = activation
# pylint: disable=arguments-differ
def construct(self, x, edge_weight, g: Graph):
"""
Construct function for GINConv.
"""
g.set_vertex_attr({"h": x})
g.set_edge_attr({"w": edge_weight})
for v in g.dst_vertex:
if self.agg_type == 'sum':
ret = g.sum([s.h * e.w for s, e in v.inedges])
elif self.agg_type == 'max':
ret = g.max([s.h * e.w for s, e in v.inedges])
else:
ret = g.avg([s.h * e.w for s, e in v.inedges])
v.h = (1 + self.eps) * v.h + ret
if self.act is not None:
v.h = self.act(v.h)
return [v.h for v in g.dst_vertex]