From the course: Advanced Graph Neural Networks

Unlock this course with a free trial

Join today to access over 24,900 courses taught by industry experts.

Setting up a graph classification model

Setting up a graph classification model

From the course: Advanced Graph Neural Networks

Start my 1-month free trial Buy for my team

Setting up a graph classification model

“

- When we train a graph neural network for the purposes of graph classification as opposed to node classification, many of the steps remain the same, but there are a few differences. The first thing is you generate embeddings for each node using the usual message passing techniques. So you perform multiple rounds of message passing. That is, you aggregate information from the neighboring nodes for every node in the graph. Once you have the updated node embeddings for every node in the graph, you aggregate node embeddings across the entire graph structure into a unified graph embedding. This is referred to as the readout layer. Once you have one graph embedding representing the entire graph, you pass that through a linear classifier to categorize or classify that particular graph. Let's set up a graph neural network to perform exactly these three steps. This neural network class is called GCN, so it's a convolutional network because we still use message passing and aggregation. That…

Contents