GINConv

class dgl.nn.pytorch.conv.GINConv(apply_func=None, aggregator_type='sum', init_eps=0, learn_eps=False, activation=None)[source]

Bases: Module

Graph Isomorphism Network layer from How Powerful are Graph Neural Networks?

\[h_i^{(l+1)} = f_\Theta \left((1 + \epsilon) h_i^{l} + \mathrm{aggregate}\left(\left\{h_j^{l}, j\in\mathcal{N}(i) \right\}\right)\right)\]

If a weight tensor on each edge is provided, the weighted graph convolution is defined as:

\[h_i^{(l+1)} = f_\Theta \left((1 + \epsilon) h_i^{l} + \mathrm{aggregate}\left(\left\{e_{ji} h_j^{l}, j\in\mathcal{N}(i) \right\}\right)\right)\]

where \(e_{ji}\) is the weight on the edge from node \(j\) to node \(i\). Please make sure that e_{ji} is broadcastable with h_j^{l}.

Parameters:
  • apply_func (callable activation function/layer or None) – If not None, apply this function to the updated node feature, the \(f_\Theta\) in the formula, default: None.

  • aggregator_type (str) – Aggregator type to use (sum, max or mean), default: β€˜sum’.

  • init_eps (float, optional) – Initial \(\epsilon\) value, default: 0.

  • learn_eps (bool, optional) – If True, \(\epsilon\) will be a learnable parameter. Default: False.

  • activation (callable activation function/layer or None, optional) – If not None, applies an activation function to the updated node features. Default: None.

Examples

>>> import dgl
>>> import numpy as np
>>> import torch as th
>>> from dgl.nn import GINConv
>>>
>>> g = dgl.graph(([0,1,2,3,2,5], [1,2,3,4,0,3]))
>>> feat = th.ones(6, 10)
>>> lin = th.nn.Linear(10, 10)
>>> conv = GINConv(lin, 'max')
>>> res = conv(g, feat)
>>> res
tensor([[-0.4821,  0.0207, -0.7665,  0.5721, -0.4682, -0.2134, -0.5236,  1.2855,
        0.8843, -0.8764],
        [-0.4821,  0.0207, -0.7665,  0.5721, -0.4682, -0.2134, -0.5236,  1.2855,
        0.8843, -0.8764],
        [-0.4821,  0.0207, -0.7665,  0.5721, -0.4682, -0.2134, -0.5236,  1.2855,
        0.8843, -0.8764],
        [-0.4821,  0.0207, -0.7665,  0.5721, -0.4682, -0.2134, -0.5236,  1.2855,
        0.8843, -0.8764],
        [-0.4821,  0.0207, -0.7665,  0.5721, -0.4682, -0.2134, -0.5236,  1.2855,
        0.8843, -0.8764],
        [-0.1804,  0.0758, -0.5159,  0.3569, -0.1408, -0.1395, -0.2387,  0.7773,
        0.5266, -0.4465]], grad_fn=<AddmmBackward>)
>>> # With activation
>>> from torch.nn.functional import relu
>>> conv = GINConv(lin, 'max', activation=relu)
>>> res = conv(g, feat)
>>> res
tensor([[5.0118, 0.0000, 0.0000, 3.9091, 1.3371, 0.0000, 0.0000, 0.0000, 0.0000,
         0.0000],
        [5.0118, 0.0000, 0.0000, 3.9091, 1.3371, 0.0000, 0.0000, 0.0000, 0.0000,
         0.0000],
        [5.0118, 0.0000, 0.0000, 3.9091, 1.3371, 0.0000, 0.0000, 0.0000, 0.0000,
         0.0000],
        [5.0118, 0.0000, 0.0000, 3.9091, 1.3371, 0.0000, 0.0000, 0.0000, 0.0000,
         0.0000],
        [5.0118, 0.0000, 0.0000, 3.9091, 1.3371, 0.0000, 0.0000, 0.0000, 0.0000,
         0.0000],
        [2.5011, 0.0000, 0.0089, 2.0541, 0.8262, 0.0000, 0.0000, 0.1371, 0.0000,
         0.0000]], grad_fn=<ReluBackward0>)
forward(graph, feat, edge_weight=None)[source]

Description

Compute Graph Isomorphism Network layer.

param graph:

The graph.

type graph:

DGLGraph

param feat:

If a torch.Tensor is given, the input feature of shape \((N, D_{in})\) where \(D_{in}\) is size of input feature, \(N\) is the number of nodes. If a pair of torch.Tensor is given, the pair must contain two tensors of shape \((N_{in}, D_{in})\) and \((N_{out}, D_{in})\). If apply_func is not None, \(D_{in}\) should fit the input dimensionality requirement of apply_func.

type feat:

torch.Tensor or pair of torch.Tensor

param edge_weight:

Optional tensor on the edge. If given, the convolution will weight with regard to the message.

type edge_weight:

torch.Tensor, optional

returns:

The output feature of shape \((N, D_{out})\) where \(D_{out}\) is the output dimensionality of apply_func. If apply_func is None, \(D_{out}\) should be the same as input dimensionality.

rtype:

torch.Tensor