A Trade-Off between Sample Complexity and Computational Complexity in Learning Boolean Networks from Time-Series Data