Since we cannot apply the raw data in the dataset to train the GNN model directly, we need to go through the following steps to convert the raw data into graphs with descriptive node features and edge features:

1. Apply noise to the trajectory to have more diverse training examples
2. Construct the graph based on the distance between particles
3. Extract node-level features: particle velocities and their distance to the boundary
4. Extract edge-level features: displacement and distance between particles

If you are not interested in the data pipeline, your can skip to the end of this section. There is a detailed explanation and visualization of one data point.