Compositional Paper Dataset

The dataset folder contains two folders - one named NGSIM and the other Artificial.

The NGSIM folder contains the CSV file containing the filtered NGSIM dataset. The dataset contains the columns: 'vid', 'eX', 'eY', 'eV', 'eA', 'lcfX', 'mcfX', 'rcfX', 'lcfY', 'mcfY', 'rcfY', 'lcfV', 'mcfV', 'rcfV', 'lcfA', 'mcfA', 'rcfA', 'lcbX', 'mcbX', 'rcbX', 'lcbY', 'mcbY', 'rcbY', 'lcbV', 'mcbV', 'rcbV', 'lcbA', 'mcbA', 'rcbA', 'Label'. The 'vid' refers to the vehicle ID we give it, 'eX', 'eY', 'eV', 'eA', are the ego/subject/lane-changing vehicle position in x  and y coordinates, its velocity and its acceleration. The remaining columns refer to the position, velocity and acceleration of the six surrounding cars relative to the subject vehicle. Here, the first letter in the column-name refers to lane of the surrounding car, 'l' for left, 'r' for right and 'm' for middle lane. The second letter 'c' refers to car. The third letter refers to the position of the surrounding car w.r.t to the subject vehicle, 'f' for front and 'b' for back. The final letter refers to the vehicle state considered, 'X' for X-position, 'Y' for Y position, 'V' for velocity and 'A' for acceleration. All surrounding vehicle values are relative to the subject vehicle values. Also, there are two files in the folder, one for the US 101 Freeway and one for the Emeryville freeway. Both datasets have to be separately loaded and combined before training.

The Artificial folder contains the 4 artificial datasets. They have their own files. Files are named according to the number of classes in the dataset. For instance, the dataset with 6 classes can be found in the folder named "Data_6_classes". Since the features in this dataset do not refer to any physical entity, they are not named. Only the 'Label' column is named and it contains the label for a dataset.

The compositional datasets can be prepared from the original datasets presented here as mentioned in the paper.