mindspore_gl.nn.ChebConv
- class mindspore_gl.nn.ChebConv(in_channels: int, out_channels: int, k: int = 3, bias: bool = True)[source]
Chebyshev Spectral Graph Convolution layer. From the paper Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering paper.
\[ \begin{align}\begin{aligned}\mathbf{X}^{\prime} = {\sigma}(\sum_{k=1}^{K} \mathbf{\beta}^{k} \cdot \mathbf{T}^{k} (\mathbf{\hat{L}}) \cdot X)\\\mathbf{\hat{L}} = 2 \mathbf{L} / {\lambda}_{max} - \mathbf{I}\end{aligned}\end{align} \]\(\mathbf{T}^{k}\) is computed recursively by
\[\mathbf{T}^{k}(\mathbf{\hat{L}}) = 2 \mathbf{\hat{L}}\mathbf{T}^{k-1} - \mathbf{T}^{k-2}\]where \(\mathbf{k}\) is 1 or 2
\[ \begin{align}\begin{aligned}\mathbf{T}^{0} (\mathbf{\hat{L}}) = \mathbf{I}\\\mathbf{T}^{1} (\mathbf{\hat{L}}) = \mathbf{\hat{L}}\end{aligned}\end{align} \]- Parameters
- Inputs:
x (Tensor) - The input node features. The shape is \((N, D_{in})\) where \(N\) is the number of nodes, and \(D_{in}\) should be equal to in_channels in Args.
edge_weight (Tensor) - Edge weights. The shape is \((N\_e,)\) where \(N\_e\) is the number of edges.
g (Graph) - The input graph.
- Outputs:
Tensor, output node features with shape of \((N, D_{out})\), where \((D_{out})\) should be the same as out_size in Args.
- Raises
- Supported Platforms:
Ascend
GPU
Examples
>>> import mindspore as ms >>> from mindspore_gl.nn import ChebConv >>> from mindspore_gl import GraphField >>> from mindspore_gl.utils import norm >>> n_nodes = 2 >>> feat_size = 4 >>> edge_index = [[0, 1], [1, 0]] >>> edge_index = ms.Tensor(edge_index, ms.int32) >>> ones = ms.ops.Ones() >>> feat = ones((n_nodes, feat_size), ms.float32) >>> edge_index, edge_weight = norm(edge_index, n_nodes) >>> feat = ones((n_nodes, feat_size), ms.float32) >>> checonv = ChebConv(in_channels=feat_size, out_channels=4, k=3) >>> res = checonv(feat, edge_weight, *graph_field.get_graph()) >>> print(res.shape) (2, 4)