How to calibrate a large-scale agent-based model?