The GNS works in two modes: one-step mode and rollout mode. In one-step mode, the GNS always makes predictions with ground-truth inputs. In rollout mode, the GNS predicts positions of particles in the next step based on its own predictions in the previous step. As a result, errors accumulate over time for rollout mode.

One Step Dataset

Each datapoint in this dataset contains trajectories sliced to short time windows. We will use this dataset in the training phase because the history of particles' states are necessary for the model to make predictions. But in the meantime, since long-horizon prediction is usually inaccurate and time-consuming, we sliced the trajectories to short time windows to improve the perfomance of the model.

Rollout Dataset

Each datapoint in this dataset contains trajectories of particles over 1000 time frames. This dataset is used in the evaluation phase to measure the model's ability to makie long-horizon predictions.

Visualize a graph in the dataset

Each data point in the dataset is a pyg.data.Data object which describes a graph. We explain the contents of the first data point, and visualize the graph.