Optimal block design models for course timetabling

Reactive Tabu Search in a Team-Learning Problem

“…These two issues also exist in our problem and we have particular interest in perfect timetables if they exist. Thus, our modeling is fundamentally different from Baker et al (2000Baker et al ( , 2002 in decision variables, objective, and constraints.…”

“…Both are computationally challenging. A substantial extension of Baker et al (2000) with a second CLP model is in Baker et al (2002). Using the modulo operation in CLP and reducing some of the symmetry caused by the definition of the decision variables, their new CLP was more efficient, but the run times for both the IP and the new CLP were still very long.…”

“…Using the modulo operation in CLP and reducing some of the symmetry caused by the definition of the decision variables, their new CLP was more efficient, but the run times for both the IP and the new CLP were still very long. For example, with 12 cohorts, five terms, and four sections, it takes 15 hours for the IP on a Sun workstation with CPLEX 6.6, and eight hours for the CLP running on a Windows PC with OPL studio 3.5 (Baker et al 2002). With the same hardware, it was much harder to solve slightly larger problems effectively.…”

“…In Baker et al (2000Baker et al ( , 2002, the objectives are to maximize total exposure, the number of total pairings for all cohorts. It is unclear, however, whether all the cohorts are treated equally, or if such a timetable is balanced.…”

“…In Baker et al (2002), much effort is made to eliminate this symmetry, which says that exchanging the numbering of any two sections in a term does not change the timetable configuration. Our definition of decision variables can avoid the definition of sections in the assignment-type modeling in Baker et al (2000Baker et al ( , 2002.…”