A Sampling-based Strategy for Distributing Taxis in a Road Network for Occupancy Maximization (GIS Cup).

We present a weighted sampling strategy for distributing a system of taxi agents on a road network. We consider a setting, in which each agent operates independently, following a prescribed strategy based on historical data. Furthermore, customer requests appear dynamically and are assigned to the closest unoccupied taxi agent. We demonstrate that in this setting a simple sampling strategy based on the spatial distribution of historical data performs well in minimizing the average time that agents are unoccupied. The strategy is evaluated on taxi trip data in Manhattan and compared to various, more complex strategies.