Hi, Dgl community, I am new learner in GNN especially focusing on heterogeneous graph. For me, I thought GNN is an important and good tools in solving some complex problems. But I found some tutorial s or journal paper or other code files, most of these are focusing on improving acc value or other evaluations. and usually the last part for one code file is show the evaluations value. but I want to use gun to solve my actual problems like make the nodes cluster or link prediction. and I hardly find a complete tutorial focusing on how can I do after training model and making evaluations.
maybe this is such a stupid question for graph deep learning, I hope someone can give me some advices. Thanks in advanced!
Making evaluations is exactly simulating how those models should behave in realistic scenarios. I think you can identify your problems first, understand it and match it with a specific task definition in graph machine learning (e.g., node classification, link prediction). Then collect data and construct a benchmark including training, development, and test splits. Try different corresponding model architectures. Results on the test split will be how well you can expect the model function in real cases.