The project contains implementations of several primal-dual subgradient methods for searching traffic equilibria in Stable Dynamic and Beckmann models. Results of experiments on Anaheim transport network are included.
The following methods are implemented:
- Subgradient method with adaptive step size [arXiv:1604.08183]
- Universal gradient method [ref]
- Universal method of similar triangles [arXiv:1701.02473].
More information about models can be found in [Nesterov-de Palma] and [Beckmann]. Anaheim_Experiments.ipynb contains code of experiments on comparison of the above methods and Frank-Wolfe algorithm (only for Beckmann model).
Convergence process on 10 000 iterations for Stable Dynamic model:
Convergence process on 8000 iterations for Beckmann model (+ Frank-Wolfe algorithm):
