Fall 1999, versions v0.40 - v0.85

At this time, Ranville was willing to duplicate her efforts and use both systems to test the worthiness of our new approach. There was some concern that seniority, especially lottery numbers, would not work to management satisfaction -- our method was not as easy to explain as the previous assignment algorithm. All the reports, provided in seniority order, have shown simulated annealing produced assignments that imitate the deterministically-ordered seniority model well enough that this potential deal-breaker disappeared as an issue.

To minimize bad assignments, Ranville and her staff have incorporated the parsing and data checking provided by the DAO into the pre-assignment data entry process.

In response to our concern that multiple runs might yield less than optimal results, Ranville agreed to try new and returning students together for the first run, and every two weeks thereafter. This was a big win as the majority of the applications would be in the initial batch.

Ranville still needed the ability to block rooms manually. Minimizing this need became the inspiration to create a constraint to effectively upgrade students automatically. Although automatic upgrading was available, there was no way to distinguish moveable students from the final manual placements Ranville made day-to-day while talking to parents, students and coaches. Automatic upgrades would have to wait.

Unfortunately for science, this first use of DAO proved so successful that Ranville quit the side-by-side testing and made the decision to use our assignment program exclusively instead. We learned some important lessons this first summer of production:

1. HMS roommate logic could ignore our suggested placements;
2. HMS had to be completely shutdown before copying its files;
3. Setting the $partial\_fill$ error too high at the beginning of the summer when there were plenty of spaces gave us unexplainable placements;
4. Though presumably helpful to the HMS assignment process, filling in choices for students who left their preferences blank on their application forms, took away some degrees of freedom from the DAO;
5. Automatically setting all non-smokers as objectors removed the possibility of giving those who actually made the designation on their application extra consideration placing them in non-smoking dorms (discussed on page ).
6. We needed a way to distinguish moveable and non-moveable assignments for automatic upgrading to work.

Running the final assignments after manual upgrading through the objective function, we found Ranville achieved a global grade of 171,948,075 where 608 students had a perfect grade, and there was 87.3% satisfaction where most students were placed in one of their requested hall. Anecdotally, the Area Coordinators noted there were fewer room change requests.

download thesis