.. DO NOT EDIT. .. THIS FILE WAS AUTOMATICALLY GENERATED BY SPHINX-GALLERY. .. TO MAKE CHANGES, EDIT THE SOURCE PYTHON FILE: .. "tutorials/models/1_gnn/1_gcn.py" .. LINE NUMBERS ARE GIVEN BELOW. .. only:: html .. note:: :class: sphx-glr-download-link-note :ref:`Go to the end ` to download the full example code. .. rst-class:: sphx-glr-example-title .. _sphx_glr_tutorials_models_1_gnn_1_gcn.py: .. _model-gcn: Graph Convolutional Network ==================================== **Author:** `Qi Huang `_, `Minjie Wang `_, Yu Gai, Quan Gan, Zheng Zhang .. warning:: The tutorial aims at gaining insights into the paper, with code as a mean of explanation. The implementation thus is NOT optimized for running efficiency. For recommended implementation, please refer to the `official examples `_. This is a gentle introduction of using DGL to implement Graph Convolutional Networks (Kipf & Welling et al., `Semi-Supervised Classification with Graph Convolutional Networks `_). We explain what is under the hood of the :class:`~dgl.nn.GraphConv` module. The reader is expected to learn how to define a new GNN layer using DGL's message passing APIs. .. GENERATED FROM PYTHON SOURCE LINES 26-47 Model Overview ------------------------------------------ GCN from the perspective of message passing ``````````````````````````````````````````````` We describe a layer of graph convolutional neural network from a message passing perspective; the math can be found `here `_. It boils down to the following step, for each node :math:`u`: 1) Aggregate neighbors' representations :math:`h_{v}` to produce an intermediate representation :math:`\hat{h}_u`. 2) Transform the aggregated representation :math:`\hat{h}_{u}` with a linear projection followed by a non-linearity: :math:`h_{u} = f(W_{u} \hat{h}_u)`. We will implement step 1 with DGL message passing, and step 2 by PyTorch ``nn.Module``. GCN implementation with DGL `````````````````````````````````````````` We first define the message and reduce function as usual. Since the aggregation on a node :math:`u` only involves summing over the neighbors' representations :math:`h_v`, we can simply use builtin functions: .. GENERATED FROM PYTHON SOURCE LINES 47-61 .. code-block:: Python import os os.environ["DGLBACKEND"] = "pytorch" import dgl import dgl.function as fn import torch as th import torch.nn as nn import torch.nn.functional as F from dgl import DGLGraph gcn_msg = fn.copy_u(u="h", out="m") gcn_reduce = fn.sum(msg="m", out="h") .. GENERATED FROM PYTHON SOURCE LINES 62-70 We then proceed to define the GCNLayer module. A GCNLayer essentially performs message passing on all the nodes then applies a fully-connected layer. .. note:: This is showing how to implement a GCN from scratch. DGL provides a more efficient :class:`builtin GCN layer module `. .. GENERATED FROM PYTHON SOURCE LINES 70-88 .. code-block:: Python class GCNLayer(nn.Module): def __init__(self, in_feats, out_feats): super(GCNLayer, self).__init__() self.linear = nn.Linear(in_feats, out_feats) def forward(self, g, feature): # Creating a local scope so that all the stored ndata and edata # (such as the `'h'` ndata below) are automatically popped out # when the scope exits. with g.local_scope(): g.ndata["h"] = feature g.update_all(gcn_msg, gcn_reduce) h = g.ndata["h"] return self.linear(h) .. GENERATED FROM PYTHON SOURCE LINES 89-95 The forward function is essentially the same as any other commonly seen NNs model in PyTorch. We can initialize GCN like any ``nn.Module``. For example, let's define a simple neural network consisting of two GCN layers. Suppose we are training the classifier for the cora dataset (the input feature size is 1433 and the number of classes is 7). The last GCN layer computes node embeddings, so the last layer in general does not apply activation. .. GENERATED FROM PYTHON SOURCE LINES 95-112 .. code-block:: Python class Net(nn.Module): def __init__(self): super(Net, self).__init__() self.layer1 = GCNLayer(1433, 16) self.layer2 = GCNLayer(16, 7) def forward(self, g, features): x = F.relu(self.layer1(g, features)) x = self.layer2(g, x) return x net = Net() print(net) .. rst-class:: sphx-glr-script-out .. code-block:: none Net( (layer1): GCNLayer( (linear): Linear(in_features=1433, out_features=16, bias=True) ) (layer2): GCNLayer( (linear): Linear(in_features=16, out_features=7, bias=True) ) ) .. GENERATED FROM PYTHON SOURCE LINES 113-114 We load the cora dataset using DGL's built-in data module. .. GENERATED FROM PYTHON SOURCE LINES 114-128 .. code-block:: Python from dgl.data import CoraGraphDataset def load_cora_data(): dataset = CoraGraphDataset() g = dataset[0] features = g.ndata["feat"] labels = g.ndata["label"] train_mask = g.ndata["train_mask"] test_mask = g.ndata["test_mask"] return g, features, labels, train_mask, test_mask .. GENERATED FROM PYTHON SOURCE LINES 129-131 When a model is trained, we can use the following method to evaluate the performance of the model on the test dataset: .. GENERATED FROM PYTHON SOURCE LINES 131-144 .. code-block:: Python def evaluate(model, g, features, labels, mask): model.eval() with th.no_grad(): logits = model(g, features) logits = logits[mask] labels = labels[mask] _, indices = th.max(logits, dim=1) correct = th.sum(indices == labels) return correct.item() * 1.0 / len(labels) .. GENERATED FROM PYTHON SOURCE LINES 145-146 We then train the network as follows: .. GENERATED FROM PYTHON SOURCE LINES 146-176 .. code-block:: Python import time import numpy as np g, features, labels, train_mask, test_mask = load_cora_data() # Add edges between each node and itself to preserve old node representations g.add_edges(g.nodes(), g.nodes()) optimizer = th.optim.Adam(net.parameters(), lr=1e-2) dur = [] for epoch in range(50): if epoch >= 3: t0 = time.time() net.train() logits = net(g, features) logp = F.log_softmax(logits, 1) loss = F.nll_loss(logp[train_mask], labels[train_mask]) optimizer.zero_grad() loss.backward() optimizer.step() if epoch >= 3: dur.append(time.time() - t0) acc = evaluate(net, g, features, labels, test_mask) print( "Epoch {:05d} | Loss {:.4f} | Test Acc {:.4f} | Time(s) {:.4f}".format( epoch, loss.item(), acc, np.mean(dur) ) ) .. rst-class:: sphx-glr-script-out .. code-block:: none NumNodes: 2708 NumEdges: 10556 NumFeats: 1433 NumClasses: 7 NumTrainingSamples: 140 NumValidationSamples: 500 NumTestSamples: 1000 Done loading data from cached files. /opt/conda/envs/dgl-dev-cpu/lib/python3.10/site-packages/numpy/core/fromnumeric.py:3504: RuntimeWarning: Mean of empty slice. return _methods._mean(a, axis=axis, dtype=dtype, /opt/conda/envs/dgl-dev-cpu/lib/python3.10/site-packages/numpy/core/_methods.py:129: RuntimeWarning: invalid value encountered in scalar divide ret = ret.dtype.type(ret / rcount) Epoch 00000 | Loss 1.9552 | Test Acc 0.1180 | Time(s) nan Epoch 00001 | Loss 1.8349 | Test Acc 0.1780 | Time(s) nan Epoch 00002 | Loss 1.6952 | Test Acc 0.3470 | Time(s) nan Epoch 00003 | Loss 1.5563 | Test Acc 0.4120 | Time(s) 0.0040 Epoch 00004 | Loss 1.4401 | Test Acc 0.5120 | Time(s) 0.0041 Epoch 00005 | Loss 1.3289 | Test Acc 0.5910 | Time(s) 0.0041 Epoch 00006 | Loss 1.2240 | Test Acc 0.6450 | Time(s) 0.0041 Epoch 00007 | Loss 1.1310 | Test Acc 0.6820 | Time(s) 0.0041 Epoch 00008 | Loss 1.0451 | Test Acc 0.7080 | Time(s) 0.0041 Epoch 00009 | Loss 0.9626 | Test Acc 0.7250 | Time(s) 0.0041 Epoch 00010 | Loss 0.8841 | Test Acc 0.7440 | Time(s) 0.0041 Epoch 00011 | Loss 0.8109 | Test Acc 0.7510 | Time(s) 0.0041 Epoch 00012 | Loss 0.7440 | Test Acc 0.7590 | Time(s) 0.0040 Epoch 00013 | Loss 0.6824 | Test Acc 0.7560 | Time(s) 0.0040 Epoch 00014 | Loss 0.6237 | Test Acc 0.7550 | Time(s) 0.0040 Epoch 00015 | Loss 0.5693 | Test Acc 0.7630 | Time(s) 0.0040 Epoch 00016 | Loss 0.5206 | Test Acc 0.7630 | Time(s) 0.0040 Epoch 00017 | Loss 0.4764 | Test Acc 0.7640 | Time(s) 0.0040 Epoch 00018 | Loss 0.4361 | Test Acc 0.7680 | Time(s) 0.0040 Epoch 00019 | Loss 0.3991 | Test Acc 0.7700 | Time(s) 0.0040 Epoch 00020 | Loss 0.3650 | Test Acc 0.7730 | Time(s) 0.0040 Epoch 00021 | Loss 0.3337 | Test Acc 0.7700 | Time(s) 0.0040 Epoch 00022 | Loss 0.3054 | Test Acc 0.7740 | Time(s) 0.0040 Epoch 00023 | Loss 0.2798 | Test Acc 0.7650 | Time(s) 0.0040 Epoch 00024 | Loss 0.2567 | Test Acc 0.7620 | Time(s) 0.0040 Epoch 00025 | Loss 0.2357 | Test Acc 0.7600 | Time(s) 0.0040 Epoch 00026 | Loss 0.2166 | Test Acc 0.7580 | Time(s) 0.0040 Epoch 00027 | Loss 0.1991 | Test Acc 0.7560 | Time(s) 0.0040 Epoch 00028 | Loss 0.1830 | Test Acc 0.7510 | Time(s) 0.0040 Epoch 00029 | Loss 0.1682 | Test Acc 0.7480 | Time(s) 0.0040 Epoch 00030 | Loss 0.1546 | Test Acc 0.7480 | Time(s) 0.0040 Epoch 00031 | Loss 0.1420 | Test Acc 0.7440 | Time(s) 0.0040 Epoch 00032 | Loss 0.1306 | Test Acc 0.7420 | Time(s) 0.0040 Epoch 00033 | Loss 0.1201 | Test Acc 0.7410 | Time(s) 0.0040 Epoch 00034 | Loss 0.1105 | Test Acc 0.7420 | Time(s) 0.0040 Epoch 00035 | Loss 0.1018 | Test Acc 0.7410 | Time(s) 0.0040 Epoch 00036 | Loss 0.0938 | Test Acc 0.7370 | Time(s) 0.0040 Epoch 00037 | Loss 0.0866 | Test Acc 0.7360 | Time(s) 0.0040 Epoch 00038 | Loss 0.0799 | Test Acc 0.7380 | Time(s) 0.0040 Epoch 00039 | Loss 0.0739 | Test Acc 0.7380 | Time(s) 0.0040 Epoch 00040 | Loss 0.0683 | Test Acc 0.7370 | Time(s) 0.0040 Epoch 00041 | Loss 0.0633 | Test Acc 0.7360 | Time(s) 0.0040 Epoch 00042 | Loss 0.0587 | Test Acc 0.7350 | Time(s) 0.0040 Epoch 00043 | Loss 0.0545 | Test Acc 0.7370 | Time(s) 0.0040 Epoch 00044 | Loss 0.0507 | Test Acc 0.7390 | Time(s) 0.0040 Epoch 00045 | Loss 0.0472 | Test Acc 0.7390 | Time(s) 0.0040 Epoch 00046 | Loss 0.0440 | Test Acc 0.7400 | Time(s) 0.0040 Epoch 00047 | Loss 0.0411 | Test Acc 0.7410 | Time(s) 0.0040 Epoch 00048 | Loss 0.0384 | Test Acc 0.7400 | Time(s) 0.0040 Epoch 00049 | Loss 0.0360 | Test Acc 0.7390 | Time(s) 0.0040 .. GENERATED FROM PYTHON SOURCE LINES 177-206 .. _math: GCN in one formula ------------------ Mathematically, the GCN model follows this formula: :math:`H^{(l+1)} = \sigma(\tilde{D}^{-\frac{1}{2}}\tilde{A}\tilde{D}^{-\frac{1}{2}}H^{(l)}W^{(l)})` Here, :math:`H^{(l)}` denotes the :math:`l^{th}` layer in the network, :math:`\sigma` is the non-linearity, and :math:`W` is the weight matrix for this layer. :math:`\tilde{D}` and :math:`\tilde{A}` are separately the degree and adjacency matrices for the graph. With the superscript ~, we are referring to the variant where we add additional edges between each node and itself to preserve its old representation in graph convolutions. The shape of the input :math:`H^{(0)}` is :math:`N \times D`, where :math:`N` is the number of nodes and :math:`D` is the number of input features. We can chain up multiple layers as such to produce a node-level representation output with shape :math:`N \times F`, where :math:`F` is the dimension of the output node feature vector. The equation can be efficiently implemented using sparse matrix multiplication kernels (such as Kipf's `pygcn `_ code). The above DGL implementation in fact has already used this trick due to the use of builtin functions. Note that the tutorial code implements a simplified version of GCN where we replace :math:`\tilde{D}^{-\frac{1}{2}}\tilde{A}\tilde{D}^{-\frac{1}{2}}` with :math:`\tilde{A}`. For a full implementation, see our example `here `_. .. rst-class:: sphx-glr-timing **Total running time of the script:** (0 minutes 0.363 seconds) .. _sphx_glr_download_tutorials_models_1_gnn_1_gcn.py: .. only:: html .. container:: sphx-glr-footer sphx-glr-footer-example .. container:: sphx-glr-download sphx-glr-download-jupyter :download:`Download Jupyter notebook: 1_gcn.ipynb <1_gcn.ipynb>` .. container:: sphx-glr-download sphx-glr-download-python :download:`Download Python source code: 1_gcn.py <1_gcn.py>` .. container:: sphx-glr-download sphx-glr-download-zip :download:`Download zipped: 1_gcn.zip <1_gcn.zip>` .. only:: html .. rst-class:: sphx-glr-signature `Gallery generated by Sphinx-Gallery `_