Add datasets with more data

[chrsmrrs]

Currently, the small number of graphs induce high variance, making it hard to compare different approach. Hence, a dataset with around 10k graphs would be good.