Machine-Learned Prediction Equilibrium for Dynamic Traffic Assignment

Authors: Lukas Graf, Tobias Harks, Kostas Kollias, Michael Markl5059-5067

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We complement our theoretical analysis by an experimental study, in which we systematically compare the induced average travel times of different predictors, including a machine-learning model trained on data gained from previously computed equilibrium flows, both on a synthetic and a real road network.
Researcher Affiliation Collaboration 1University of Augsburg 2Google {lukas.graf, tobias.harks}@math.uni-augsburg.de, kostaskollias@google.com, michael.markl@student.uni-augsburg.de
Pseudocode No The paper describes algorithmic steps in paragraph form but does not include a structured pseudocode or algorithm block.
Open Source Code Yes The code of our simulations is publicly available in (Graf et al. 2021b). ... (Graf et al. 2021b). Simulation of Dynamic Prediction Equilibria. https://github.com/schedulaar/dynamic-prediction-equilibria. Accessed: 2021-12-15.
Open Datasets Yes The second graph is the road map of Sioux Falls as given in (Le Blanc, Morlok, and Pierskalla 1975) which is commonly used in the transport science literature. ... The third graph is the center of Tokyo as obtained from Open Street Maps (Open Street Map contributors 2017).
Dataset Splits No The paper mentions generating 'training data' for the machine learning predictor but does not specify explicit training/validation/test dataset splits (e.g., percentages, sample counts, or references to predefined splits).
Hardware Specification Yes T comp avg denotes the average computation time for computing a single ε-DPE on a single core of an Intel Core i7-3520M CPU at 2.90GHz.
Software Dependencies No The paper does not specify version numbers for key software components or libraries used (e.g., 'Python 3.x', 'PyTorch 1.x').
Experiment Setup Yes In our implementation, we assume that shortest paths as predicted at some time θ stay shortest paths for a certain time interval [ θ, θ +ε) and agents compute new shortest paths every ε time units resulting in ε-DPEs... The features used to train the model are 10 observations of the past queue length of the edge and of neighboring edges.