Home
BEAMPyOpt : Calibrating large-scale agent-based transport simulations
BEAMPyOpt is a calibration tool for BEAM modeling framework. The purpose of developing BEAMPyOpt is two-fold: to quickly evaluate the extent of the unknown search space for optimizing the specific black-box function and to find a global optimum of the black-box function from a narrowed search space. BEAMPyOpt is capable of optimizing 8 hyperparameters where each hyperparameter affects the simulation in complex ways.
BEAMPyOpt runs on a distributed system in parallel. We have implemented two algorithms in this package which is suitable to optimize the Mode Choice Model based on multinomial logit regression. One of the implemented algorithms is Search Space Control to find the range of the search space where the global optima lie. Another algorithm combines Bayesian Optimization and HyperBand to efficiently search for well-performing configuration in the narrowed search space.
BEAMPyOpt can restore the optimization information from an existing scenario and restart later to guide the optimizer to search in areas where the model has historically performed well. Our approach can be extended to other objective functions with more input parameters. This improvement will make it easier to calibrate BEAM in new cities.
(Images are viewed better on a white background.)
BEAMPyOpt Software Architecture