Questions about the datasets: how to split the training and testing sets in dynaernn?

I have read your paper (reference 6), and i am wondering if you might provide more details about how to evaluate the impact of length of graph sequence supplied to the models during training on its performance. Thank you.