... system[*]
http://www.cbord.com/ 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... importance.[*]
An as-yet-unreleased rewrite of HMS reportedly allows the user to adjust the importance of the roommate rule. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... numbers'[*]
Known internally as lottery number, it is the number stamped on the student's application when their deposit is paid. New and returning students have separate accumulators -- lower is better. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... floors.[*]
Allison, S. Personal communication, Spring 1999. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... empty).[*]
Other notions of a `move' could be defined (e.g. a group together, or splitting a suite). 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... list.[*]
The number of virtual beds is a configurable option with a minimum of one room (six beds) per gender. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...figlandrand.[*]
To reduce clutter, an arbitrary 25% of the histogram data is plotted. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... gender[*]
Gender consensus applies to undesignated male or female rooms. The gender attribute is generally predetermined at UNM (see page [*]). 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... seniority-lottery-weight[*]
The $slw$ is the number of consecutive semesters as a resident in the dorms multiplied by the seniority weight, plus the lottery benefit (see page [*]). 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... grade.[*]
Worthy of note is the significance of the maximum number of beds in a room (a constant) to the worst case complexity of the group, gender, and smoking student-consensus constraints across suites. For an intermediate grade calculation, the complexity in the worst case, based on consensus_across_suites = # rooms_in_suite * # students_in_room * # roommate_requests_per_ student * # rooms_in_suite * maximum_beds_in_room, is $O(b^{5})$, where $b$ is the maximum number of beds in a room. Since MAX_BEDS is a constant, and a constant raised to a power is constant, the complexity is still $O(1)$. At UNM, the maximum number of beds per room, found in the apartment-style dorm, is six. Furthermore, suites comprise approximately 46% of the total bed space. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... students.[*]
Doubling the seniority weight to 2000 produced more defendable results, according to Ranville's analysis in early Fall 2000 tests. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... billion.[*]
In limits.h, ULONG_MAX = 4,294,967,295 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... separately.[*]
Simulated Annealing in Dorm Assignments, CS451 Project, October 1998. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... attempts.[*]
The choice of 1000 for the history buffer is questionable. Arguably, the size of the history buffer should be adjusted to the phase, the number of students, the number of beds, or both. With the accept-based approaches (sa3 & sa4), when the phase is 2000, there is complete turn around in the 1000 cost changes used to calculate the standard deviation. As the phase is reduced to less than 1000 in sa3, there is possible and probable overlap in the history at these later decision points. With the attempt-based approach (sa2), all the zero difference attempts are included in the history, providing little possibility of overlap. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... default.[*]
We found a value of five for the TEMP_DIVISOR is better, though slower. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... adjustment.[*]
The delay can be extended by setting the value of the DECIDE_TO_COOL configuration option (zero by default) higher. We used a value of one, which gave us at least three phases at each temperature. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... SIZE="-1">(NOBETTER).[*]
The number of rounds can be pre-determined with the LOOP_CONTROL option. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... heat[*]
``Specific heat is a measure of the variance of cost (or energy) values of states at a given temperature'' [EFC98]. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... SIZE="-1">DAO[*]
The UNIX utility gprof was included as a Makefile option. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... process.[*]
Flex and Bison UNIX utilities are used to parse each of the input data files, providing extensive data checking. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... environment.[*]
The HMS auto-assignment batch could have been backed-out and another configuration imported and re-assigned in about five hours, but by then too much time had been invested in manual upgrades that would have been lost by reversing the batch. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... better.[*]
During `Move-In-Week', `self-upgrading' replaced auto-upgrading as evidenced by a poster announcing: DOUBLES AS SINGLES -- MOVE-IN SPECIAL -- Rooms in Coronado and Santa Clara -- Available for Immediate Occupancy (six male rooms & six female rooms). 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... step[*]
For RANDOMIZING_PHASE times (default 100,000), we randomly select a student and a bed and switch them with a 50% probability, like running at a high temperature. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... rank[*]
The mean rank allows for varying sample sizes: A group's rank sum is divided by its sample size. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... others[*]
Iterated next-descent hillclimbing did, however, improve the score of an average random state (r) by 25%. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...rldhc[*]
The random lateral-descent hillclimbing took as long as the algorithmic scheme intentionally to give it the fairest opportunity to find a good configuration. As hillclimbing was not our focus, we offer no data regarding its performance in fewer attempts. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... middle.[*]
The Metropolis variations of the table-driven and reheating schemes behaved similarly to our simulated annealing versions in number of attempts. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... 2001.[*]
Sultemeier, J. Personal communication, September 12, 2000. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... Function[*]
This description corresponds to the last time it was modified in DAO version v1.12, when the different_bed constraint was added. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... follows:[*]
Default values are zero unless specified (this policy was adopted in version v0.84). 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.