Вы находитесь на странице: 1из 18

Kyle T.

Kowalczyk
PSU Student, Class of 2019
513 West College Avenue, Apartment #6
State College PA 16803 The Pennsylvania State
University’s Office of the
December 11th, 2017
Registrar
Dr. Robert A. Kubat
Pennsylvania State University Registrar

Todd Clouser
Director of IT

Dear Gentlemen,
I submit herewith, a recommendation for potential
improvements to the course selection process, here at
PSU. As a student that has personally experienced the
use and results of the course selection process, I have a
special interest in improving this most detailed and stressful process.
Kyle Kowalczyk
This report examines the current system, and offers a solution through| revision
Decemberof
11,the
2017 |
course
selection algorithm. Currently, the process is solely, hierarchal based. And while, this does offer
a degree of fairness to the student body, it is one-dimensional. Most freshman and sophomores
focus on the lower level introductory classes. Once they complete them, they still have to
skirmish for the upper level 300 and 400 classes. This effectively, negates any utility of a
hierarchical system, and leads to longer graduating times.

Course Selection
With the recent implementation of an inclusive student administrative system (Lionpath), it is
only fitting that we update the tired course selection system.

Algorithm
I would like to express my sincere gratitude to Timothy Kowalczyk, a principle member of the
“AL-3
plate of frustrationengineering
Technologies’ with a side staff,
order for
of angst”
algorithm consultation.

Amendment
I would be most grateful if you could spare time to review this recommendation report. If you
have any questions, after reading the report, I would be most happy to answer any question that
would arise.

Respectfully,
Kyle Kowalczyk

Enclosure: Recommendation Report


Kyle T. Kowalczyk
PSU Student, Class of 2019
513 West College Avenue, Apartment #6
State College PA 16803

December 11th, 2017

Dr. Robert A. Kubat


Pennsylvania State University Registrar

Todd Clouser
Director of IT

Dear Gentlemen,
I submit herewith, a recommendation for potential improvements to the course selection process,
here at PSU. As a student that has personally experienced the use and results of the course
selection process, I have a special interest in improving this most detailed and stressful process.

This report examines the current system, and offers a solution through revision of the course
selection algorithm. Currently, the process is solely, hierarchal based. And while, this does offer
a degree of fairness to the student body, it is one-dimensional. Most freshman and sophomores
focus on the lower level introductory classes. Once they complete them, they still have to
skirmish for the upper level 300 and 400 classes. This effectively, negates any utility of a
hierarchical system, and leads to longer graduating times.

With the recent implementation of an inclusive student administrative system (Lionpath), it is


only fitting that we update the tired course selection system.

I would like to express my sincere gratitude to Timothy Kowalczyk, a principle member of the
L-3 Technologies’ engineering staff, for algorithm consultation.

I would be most grateful if you could spare time to review this recommendation report. If you
have any questions, after reading the report, I would be most happy to answer any question that
would arise.

Respectfully,
Kyle Kowalczyk

Enclosure: Recommendation Report


Table of Contents
Title Page ....................................................................................................................................................... i
Letter of Transmittal ........................................................................................................................ ii
Table of Contents ............................................................................................................................ iii
Executive Summary ........................................................................................................................ iv

Introduction ................................................................................................................................................... 1
Stable Marriage .......................................................................................................................................... 2-4
Genetic Algorithm ..................................................................................................................................... 5-7

Appendix A – Annotated Bibliography Table of Contents ............................................................. A-1 – A-2


Appendix B – Algorithm Discussion .............................................................................................. A-3 – A-5

List of Figures
Figure 1 – Stable marriage ............................................................................................................................ 2

Figure 2 - Explanation of Stable marriage preferences..................................................................................... 4

Figure 3 - Genetic Algorithm Logical overview .............................................................................................. 8

iii
Executive Summary

Highest quality education at the lowest price. This is a goal that will, ultimately serve every
stakeholder at the Pennsylvania State University. The current issue with Penn State’s course
scheduling system, is that a hierarchical system wastes resources.
In this report we explore two main revisions to the current credit based algorithm. These are the
Stable Marriage Algorithm, and Genetic Algorithm.
The Stable marriage algorithm aims to match a class seat with a student so that they have a stable
relationship (the student completes the course). However, this algorithm is very inelastic. Here
lies its strengths and weaknesses. Due to its rigidly and the fact that there are many different
conflicting variables, this algorithm will be difficult to implement in a realistic manor.
The Genetic Algorithm is flexible. It shines where stable marriage does not. All constraints and
variables are able to be defined and applied to the fitness function(s). This allows for flexibility.
And, ultimately, for us to find the most efficient set of student schedules (so that students are
least likely to drop out of their classes and all class seats are filled).
While, the genetic solution does require a higher up-front investment, (server processing power
and system development), the long-run returns will be well worth it. Not only will it differentiate
our university on a technological stage, but it allows all of our current resources to be
maximized. Currently, there is a large waste of professor capital when students don’t stay in a
class (drop), or fill up the class seats. To make things worse, there is no shortage of student
dissatisfaction (even with a hieratical credit based system). No party is completely satisfied with
the current system.
With a new genetic, course selection algorithm, students, faculty, and the university will be
better off. No additional faculty will have to be hired. More students will stick with their classes,
and as a result, more students can be admitted to the university.
By optimizing our current resources with a genetic algorithm, the university can expand greatly.
All of these benefits without major cost to the university or the customer. Waste is eliminated, as
we invest in the future of our university.

iv
Introduction
The Pennsylvania State University serves 82,400 undergraduate students, in total. Around 42,000
of them attend the university in University Park, PA.All of these students have access to and
currently use an all-inclusive student account management system called Lionpath. This multi-
year initiative began in 2014, and as of fall of 2017, replaced its predecessor Elion. The main
advantage between the two, being that all aspects of the student account at the university would
be accessible and controllable from one singular system.
Lionpath was cleaner looking and consolidated many different university functions (e.g.
Financials, Academics, course scheduling, etc.). However, it changed very little about the course
scheduling process, itself (wait-listing changes from lottery system to hierarchal). The main
function of Elion was for it to be a platform for students to self-schedule their classes for the
coming semester. This was always, and will always, be a main concern for the student body.
To maximize their best chances of success, students take different classes depending on self-
defined criteria; things like teacher ratings and requirements for their major. Thankfully, humans
are extremely efficient creatures, who are all simply just trying to maximize the value they get
out of each situation. So, naturally, there is a lot of overlap between different classes that
students want. The current solution to this limited resource problem, is a credit-hieratical,
scheduling system.
The end goal is to have the best quality education. Every stakeholder can agree with that. The
problem with a hieratical system is that many, resources are wasted (namely teacher capital).
This could be due to many things like students dropping out of a class they weren’t prepared for
or were planning to take. Whenever a class is not fully filled up, that is a waste of the
university’s resources. With issues with overcrowding at main campus, if all classes were filled
and dropping rates were lower, more students could be admitted to the school.
Additionally, there is the student’s perspective that was touched on earlier. When there is a
limited supply of classes that underclassman want, Underclassmen are (statically and
significantly) looking for intro level courses and general education classes, foremost. They will
not get the schedule that they want because the scheduling window is later due to credit
seniority. Upperclassmen, need their higher 300s and 400s to try and graduate within their goals.
The issue with this hierarchal system is that too many people have the same goal. All of these
upperclassmen (and underclassmen), in their majors, want certain professors or classes. So, even
though they have the seniority to register first, they are still unable to get the schedule that they
want. The seniority system is causing students longer to get their degrees, and unnecessary
frustration among students.
While the obvious solution is to hire professors in quantity and quality, there is no need to
sacrifice even more resources and capital. I propose that the class selection algorithm is
optimized so that class-dropping rates decrease, enrollment numbers inflate, and professor
recruitment holds constant (or even decreases). (Hoovers)

1
Stable marriage

Introduction
Humans are extremely resourceful
creatures. A core principal of economics is
that we are logical creatures that make
decisions solely based around what will
give the greatest benefit to each individual.
That is to say, we naturally will optimize
our life by choosing the best options that are
presented to us. The same assumptions hold
true when talking about students and course
selection. Under the assumption that
information is free flowing, the student will
automatically pick the classes that will give
him the best chances of completing their Picture 1 – Stable marriage - Retrieved from www.mercatornet.com
goals. The stable marriage algorithm takes
this into consideration and attempts to assign each student with a stable class schedule (with few
class drops).

Common student goals:


(long term) (Short term)
-Graduate within four years -Secure desirable classes (due to
-Attain a high G.P.A. teacher, subject matter, location etc.)

Why Stable marriage?


Unfortunately, due to so many students with similar goals, there are a very limited number of
desirable classes, and therefore ultimately student schedule satisfaction. More often than not,
underclassman will get dreaded “Eight AMs” because they couldn’t secure the class schedule
they had originally chose a to be the most efficient to them as an individual. So, is there a way to
fairly divide up a limited amount of resources, regardless of class standing?

The Stable marriage algorithm takes the moral stance that


seniority should not be a determinant of who gets to choose what class that they want. After all,

2
everyone at a university is there because they
wish to better themselves to some degree, and had
to sacrifice time money and energy to be there.
Why should a seniority system prejudice against How does it work?
students because they are newer at the university?
The morally correct ideology, in this case, is
simply that the resources should be spread out
In this case, we will define the parameters as
evenly to everyone, objectively. follows. Every student registered (defined as n) at
the university, with a set list of preferences for
Classically, the Stable marriage algorithm was classes (defined as m) are men the looking for a
designed to answer the question “Is it always stable marriage. The classes which the students
possible to arrange stable marriages (given that desire (class number), are the women and each one
has a limited number of seats they can marry (this
there are n amount of people, with preferences for is class size - q). The problem that the algorithm
each other)? The answer is yes. There is an looks to solve is that the student n, with a list of m
arrangement that will provide everyone with preferences (let [m = x1, x2, x3, x4, x5, x6, x7, x8, x9 ],
happy, stable marriages. (See attached appendix and each xz represents the prioritized class of nine
for brief introduction to the Stable Marriage preferences that each n prefers for their schedules).
For our purposes (to promote fairness), we will not
Algorithm) give q, its own set of preferences.
[But, it is important to note that this is very simple
Criteria to do and opens a lot of possibilities. For example,
we can make q prefer students with higher grade
Advantages: point average. This could provide an incentive for
• Optimized solution where both student higher graduating rates, and lower class drop rates,
and university goals are met for students in classes that they need for graduation.
Less controversially, seniority can be integrated
• University gets good publicity in into the q preferences. The difference this time, is
attempting to improve course selection that student input is taken into consideration, so
system there will be a higher distribution of filled classes
due to satisfaction.]
Rigidity is not always a bad thing. Because
classes are evenly distributed to the best *note* Traditionally, the preferences defined by q
are imposed by seniority. However, because even if
optimized solution among all students in the
a student is satisfied with their class schedule, this
university system, the amount of extra open seats does not guarantee that there will be no anomalies
is reduced to a completely known number. This is (students who will like to switch their classes or
a revolutionary concept as, admissions will no drop). Therefore, it’s possible that undesired waste
longer have to guess when it comes to what size will occur in addition to unintended preferences
from q.
should the next year of students be. With the
implementation of this algorithm there is no The algorithm will run past each student, trying to
uncertainty. Admission will be able to admit the line them up with their preferences against all other
exact number of students so that every seat at variables in the system (things like class times
conflicts, prerequisites, and student preferences).
every class offered at the university is filled. The algorithm runs through each student one at a
time and assigns stable class spots to each student.
Disadvantages: It will run at least nine times, and until is assigns
• Computationally complex – May take each student with a stable schedule. Thanks to the
Stable Marriage Class Selection Algorithm, even if
much computation resources a student doesn’t get their first six choices as their
• Difficult to implement schedule, each and every student, ends up with a
class schedule that is stable, to them.

3
• Somewhat difficult to explain / operate for students
The main flaws in this algorithm, lies in its architecture. By definition, this algorithm is rigid.
Once the parameters are set, it will be difficult to adjust student schedules at all. Additionally, it
nullifies the current add/drop grace period system. Students will have the option to take their
given predefined schedule and that is all. If they want drop the class that is also fine. No extra
classes past the average of 16 will be standardized. If students want to take extra classes, they
will not be able to because (assuming an algorithm operating efficiency of 100%), there is no
room. Finally, there is complications with special cases (like student athletes, who often need an
additionally customized schedule due to training requirements). With many different conflicting
variables, this algorithm will most likely be difficult to implement effectively. (Gale, 1962)

Figure 1 - Explanation of Stable marriage preferences. Retrieved from Hoang.

4
Genetic Algorithm
Introduction
This report embraces the fact that humans have optimized our environment and chose the best
suited path (that has the highest chances of success). To our current knowledge, there is no more
intelligent or complex being in the universe, than the human. Is this a testament to the greatest
tool for optimization? Is there a more efficient way to overcome constrains? Our mere existence
answers with resonating “no”. The key to our success lies in a cruel process that we call
evolution. Evolution takes many generations to occur. Simply put, the better suited organism gets
to pass on its genetic information. With the advent of supercomputing and cloud services, it is
possible to build an algorithm that responds to our constraints, evolves, and produces a
satisfactory solution. Its name is the Genetic Algorithm.

Why Genetic?
The genetic algorithm has been used successfully in search heuristics, for many years. Once
computers became fast enough to process things affordably for the masses, its use exploded.
Genetic algorithms, or evolutionary algorithms can be very efficient. Instead of pointlessness in
calculating many different possible variations, the algorithm changes and prioritizes
things(parameters) that are more important (by tweaking the fitness function). This allows us to
arrive at satisfactory solutions with less computational waste.

What is the Genetic Algorithm?


The Genetic Algorithm (GA) is modeled after the process of biological life and the success of the
most desirable outcome for the biological species as a whole - evolution. The attributes or
characteristics of the individual decisions to be made and encoded in various potential solutions
much like biological chromosomes to determine individual entity characteristics. The GA is an
enhanced and formalized generalization of a 1970s’ early computer game called Conway’s Game
of life that represented graphical cellular evolution. (Conway’s Game)
The process begins with the encoding of attributes onto candidate solutions instances. These
instances are then evaluated for fitness against the goals of an optimum candidate. If the Fitness
does not meet some defined success criteria, then then phase two of the algorithm begins. A
selection function is used to promote some of the candidate solutions to the next round. Those
candidates surviving this round are then subjected to two additional functions; namely crossover
and mutation. Crossover mimics parental contribution of two parents to a single offspring where
some percental originates from one parent and the remainder from the other parent. Mutations
result from the random reassignment of random characteristics in each candidate solution. This
process then repeats back to the Fitness calculation then crossover and mutation until the exit
criteria is reached or a limit on closure is attained.

5
Advantages
• Generated solutions have high of fitness to desired goals
• Well understood implementation
• Criteria is easily encoded in candidate solution
• Commercial solution (like EVOLVER ) are available

The advantages of the Evolution algorithm, over the Stable marriage algorithm, are comparable
to their human given worth. While Stable marriage, is important to humans in finding their
immediate (immediate - in terms to timescale) happiness. Evolution allowed for our success as a
race. That is to say, where the Stable marriage algorithm has many immediate and practical
challenges to overcome (before it can be implemented), the Genetic solution, does not. Again,
this lies in its architecture. Where the stable marriage was frigid, the evolution algorithm, is
always changing. Constraints will be pre-programmed into the fitness function, so that they no
longer become conflicting. Not much administrator (Academic advisor) override, will be
required for special cases (like student athlete). As the name suggests, evolution’s strength lies in
its innate ability to overcome conflict and change to the environment.

Disadvantages
• As number of parameters increase, computational burden increases exponentially
• May achieve a local optimal and a non-optimum solution

While, the Genetic algorithm possess the ability to overcome many constraints at much less
resources than brute force calculations, there are still some constraints. Server power will the
major cost in this endeavor. Because every students course schedule is treated as its own
population, the algorithm runs for each student separately, assigning each one a class that fulfills
the variables (fitness test), and then systematically runs over and over until all students receive
their schedules. With over 100,000 students in the PSU system, and 1,000s of classes, it will take
a lot of computational power to come up with an optimal solution for every case. Luckily, we are
in the age of computers, and recent mass-marketed solutions like Amazon web services and
cloud computing allow something like the genetic algorithm to become financially feasible.

Conclusion

Another huge benefit for of implementing the genetic algorithm into the course selection process,
is how easy it will be to convert the pre-course selection process into a genetic algorithm based
selection process. This software will be easy and cheap to implement due to the fact that genetic
algorithmic systems already exist (and are being used) for administrative class scheduling. We
can skip the prototyping, testing, and implementation processes. By adding this system to our
new course selection algorithm amendment, we will completely revolutionize and automatize,
the course selection process. Both front end, and backend, goals, will be met.

While, the solution does require a bit of up-front investment, (going towards server processing
power, and system development), the long-run returns will be well worth it. Not only will it
differentiate our university on a technological stage, but it allows all of our current resources to

6
be maximized. Currently, there is a large waste
of professor capital when students don’t stay in
How does it work? a class (drop), or fill up the class seats. To make
things worse, there is a large amount of student
dissatisfaction even with a hieratical credit
based system. No party is completely satisfied
The population consists of a single individual. with the current system.
Therefore, by having a single schedule we can avoid
problems such as checking that class sections were not With a new genetic, course selection algorithm,
repeated or got lost on the schedules (when
performing the mutation operations). The single
students, faculty, and the university will be
schedule used here was randomly generated using better off. No additional faculty will have to be
information about class size, class sections, hired. More students will stick with their
instructors, classrooms, pre-requisites, class times, classes, and as a result more students can be
and student preferences. Data is used from the course admitted to the university. By optimizing our
listing at the Pennsylvania State University’s
Schedule of courses for the next coming semester (fall
current resources with a genetic algorithm, the
2018). Courses are commonly taught either on university can expand greatly. All of this
Monday, Wednesday, and Friday, or on Tuesday, and without major cost to the university or the
Thursday. Therefore, to make our program work with customer.
this model, we subdivide the schedule into two sub-
schedules called MWF-schedule and TTH-schedule.
Selection will be performed after the schedule is
altered by mutations.
The fitness function for the mutated schedule will
then be evaluated. Selection consists in selecting
either the mutated schedule or the schedule before the
mutations. If the mutated schedule had a higher fitness
value that the unmutated schedule, this new schedule
will be selected. Otherwise, the unmutated schedule
was kept. The program repeated this process as many
times as needed. Some of these mutations were
actually crossover operations when the population
consists of more than one schedule.
After the algorithm finds a sufficient matched based
off of the fitness function criteria, it moves on the next
student in the pool. The algorithm draws its
population from every student in the Penn state
system and all of the possible class seats from the
schedule of courses. Once a seat is secured by the
algorithm, the seat is systematically placed into
superposition, until every previous student position
hierarchy, is filled. Once this occurs, the seat is
confirmed and the algorithm continues onto the next
student schedule level. As this process occurs, the
system crosschecks that there are no time or
perquisite, conflicts, against each induvial class tree
schedule.

7
Figure 2 - Gentic Algorithm Logical overview - Retrieved from www.genetic.io

8
Works cited
Conway's Game of Life. (n.d.). Retrieved December 05, 2017, from
http://web.stanford.edu/~cdebs/GameOfLife/

Coughlan, S. (2014, September 16). What makes a global top 10 university? Retrieved December 03,
2017, from http://www.bbc.com/news/business-29086590

Ellongsen, K., & Penaloza, M. (n.d.). A Genetic Algorithm Approach for Finding a Good Course
Schedule. Department of Mathematics and Computer Science South Dakota School of Mines and
Technology. Retrieved December 02, 2017, from
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.9.9600&rep=rep1&type=pdf

GACS: Genetic Algorithm Class Scheduler. (n.d.). Retrieved December 03, 2017, from
http://gacs.sourceforge.net/

Gale, D., & Shapley, L. (1962). College Admissions and The Stability Of Marriage. The American
Mathematical Monthly, 69(1), 9-15. doi:10.21236/ad0251958

Hoang, L. N. (n.d.). Marriage Assignment Problem and Variants. Retrieved December 05, 2017, from
http://www.science4all.org/article/marriage-problem-and-variants/

Hoover's Inc. (2015). The Pennsylvania State University. Retrieved November 27, 2017 from Hoover's
database.

Lucci, S., & Knopec, D. (2016). RedmondAI_Ch12_2nd_ed_Genetic [PPT]. Mercury Learning and
Information.
Retrieved from La Salle University's CSC 456-01 Artificial Intelligence taught by Dr. Redmond

Many successful gay marriages share an open secret (2014, June 04). Retrieved December 10, 2017, from
https://www.mercatornet.com/family_edge/view/14166

Penn State University Park Academics - CollegeData College Profile. (n.d.). Retrieved December 04,
2017, from https://www.collegedata.com/cs/data/college/college_pg04_tmpl.jhtml?schoolId=59

Stable matching: Theory, evidence, and practical design. (2012). The Royal Swedish Academy of
Sciences, (THE PRIZE IN ECONOMIC SCIENCES), 1-5. Retrieved December 1, 2017, from
https://www.nobelprize.org/nobel_prizes/economic-sciences/laureates/2012/popular-
economicsciences2012.pdf.

Tremaglio, K. (2014, February 06). LionPATH celebrates official launch of project. Retrieved December
01, 2017, from http://news.psu.edu/story/302955/2014/02/06/administration/lionpath-celebrates-official-
launch-project

9
APPENDIX A – ANNOTATED
BIBLIOGRAPHY

Stable matching: Theory, evidence, and practical design. (2012). The Royal Swedish Academy of
Sciences, (THE PRIZE IN ECONOMIC SCIENCES), 1-5. Retrieved December 1, 2017, from
https://www.nobelprize.org/nobel_prizes/economic-sciences/laureates/2012/popular-
economicsciences2012.pdf.

This explains the work of Lloyd Shapley and Alvin Roth, who were awarded the noble prize in
Economic Sciences in 2012 for their work in applying a Stable matching theory to assignments of
new doctors to hospitals, students to schools, and human organs for transplant. The paper was a
great introduction before more technical papers concerning the Gale-Shapley Algorithm. It
provided a very useful economical background to the application and implications of using a
stable matching approach to real world issues. The problem with a lot of academic, technical
papers is that they lack the ability to translate an abstract idea into practical, real world,
applications. Theoretical is not too concerning to those of us living in actuality.

Gale, D., & Shapley, L. (1962). College Admissions and The Stability Of Marriage. The American
Mathematical Monthly, 69(1), 9-15. doi:10.21236/ad0251958

This is the original paper from D. Gale and L.S. Shapley that fathered the Gale-Shapley
Algorithm. The paper goes into detail about how the algorithm works and how it was originally
formulated for the process of matching students to their desired colleges, so that they had a stable
choice. The issue was that there were too many variables when considering students and colleges
as matches. Each student had their own college that they considered as their first choice.
Additionally, there were a limited number of seats available for the college to offer to the
students. Other parameters from the colleges included whether the applicant has applied to other
colleges, how he ranks within the colleges which he has applied, and which colleges will offer to
admit him. The question was, “Is there a satisfactory combination of students to their college
choices so that everyone is satisfied with their college choice; both the college and student?” The
answer is yes; there does exist a stable marriage combination for them all. This paper does a good
job describing the specific details of the algorithm and gave a relatable example to where I could
craft my own algorithms architecture. The main difference between ours is that my stable
schedule algorithm must be repeated multiple times (nine).

A-1
Ellongsen, K., & Penaloza, M. (n.d.). A Genetic Algorithm Approach for Finding a Good Course
Schedule. Department of Mathematics and Computer Science South Dakota School of Mines and
Technology. Retrieved December 02, 2017, from
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.9.9600&rep=rep1&type=pdf

This paper offered a good technical example which was similar to my goal. There are a lot of
academics, and even businesses, that are utilizing the genetic algorithm to schedule the classes
themselves. This paper considers variables like (physical) minimum number of seats, audiovisual
equipment availability, and faculty teaching preferences. It then takes the constraints and talks
about the architecture and variables that a genetic algorithm must contain in order to produce
something similar to Penn state’s Schedule of courses. The process consists of randomly
generating a single schedule for an individual, which is then mutated randomly. A fitness function
is established (that is based off the constraints) and then the mutation is either kept or discarded
based off of it. This process can be repeated until desired results. This is the work done by the
registrar before students can look at the schedule of courses. It also represents another use of the
genetic algorithm that can be integrated into the course selection process in the future.

Lucci, S., & Knopec, D. (2016). RedmondAI_Ch12_2nd_ed_Genetic [PPT]. Mercury Learning and
Information.
Retrieved from La Salle University's CSC 456-01 Artificial Intelligence taught by Dr. Redmond

This 120 slide PowerPoint reviews every single technical piece of information needed to write a
genetic algorithm (GA). It begins with explaining how the problem must be defined, along with
the constraints and optimum criteria. Once that is done the problem domain must be represented
as a chromosome. Next, we must define the fitness function that will weigh the mutations of the
chromosomes. Then, the genetic operators are constructed. Finally, the Genetic algorithm is run
and its parameter’s tuned. Realistic GA’s will use 100 to 3,000 chromosomes. There are
numerous models for selection. These include roulette wheel, elitist, scaled, increasing, and
tournament, selection. These all involve changes to the fitness function during the evolution and
selection process. This PowerPoint also explores how genetic algorithms can transition into
genetic programming; where the main goal is to systematically optimize the software over time.

Hoover's Inc. (2015). The Pennsylvania State University. Retrieved November 27, 2017 from Hoover's
database.

Hoover’s overviewed the university as a business. It showed information concerning the


academic industry as a whole, and specifics relating to The Pennsylvania State University (like
faculty numbers, students, and University park vs the other campuses).

A-2
APPENDIX B - ALGORITHM
DISCUSSION

1 The “First Come – First Serve” Algorithm


A simple lottery algorithm is currently selected method of matching students with available seats
in a specific teacher’s class. This is the current method of course matching at PSU.

The method of selection is as follows:


1. The registrar publishes a list of classes that will be held in the net course selection period.
Within each “course” is also the number of seats that are available to students wishing to
attend that class.
2. The student creates a list of candidate courses. A course is defined to be a particular
course (i.e. ENG122), at a particular time with a particular teacher.
3. At a specific, pre-notified time, a list of available classes for each student that is
permitted to select courses is now available for selection and commitment. Various
factors influence when each student can begin selection: Student/Athlete and
accumulated credit hours are some these factors.
1. This process is repeated until all student groups have competed the time window of selection.

Subsequent to first round selections, students can opt to waitlist for course slots that may become
available in the future.

Advantages:
• Easiest to implement
• Easiest to explain to students

Disadvantages:
• Most of the Admin/Teacher goals are not met
• No prioritization of student goals

A-3
2 The Stable Matching Algorithm
The stable matching algorithm is also known as the Gale–Shapley algorithm. It was designed to
match a cooperative set of paired and equal parties to form optimal match that maximizes goals
across both parties’ interests. For our course selection problem at PSU, each student would
image the man in the below example and each available seat in a teacher’s class would image the
women. It is best explained in the following marriage proposal / response scenario.

Gale and Shapley analyzed matching at an abstract, general level. They used
marriage as one of their illustrative examples. How should ten women and ten men be
matched, while respecting their individual preferences? The main challenge involved
designing a simple mechanism that would lead to a stable matching, where no couples
would break up and form new matches which would make them better off. The solution –
the Gale-Shapley “deferred acceptance” algorithm – was a set of simple rules that
always led straight to a stable matching.

The Gale-Shapley algorithm can be set up in two alternative ways: either men
propose to women, or women propose to men. In the latter case, the process begins with
each woman proposing to the man she likes the best. Each man then looks at the different
proposals he has received (if any), retains what he regards as the most attractive proposal
(but defers from accepting it) and rejects the others. The women who were rejected in the
first round then propose to their second-best choices, while the men again keep their best
offer and reject the rest. This continues until no women want to make any further
proposals. As each of the men then accepts the proposal he holds, the process comes to
an end. Gale and Shapley proved mathematically that this algorithm always leads to a
stable matching.

Advantages:

• Optimized solution where both student and university goals are met
• University gets good publicity in attempting to improve course selection system

Disadvantages:

• Computationally complex – May take much computation resources


• Difficult to implement
• Somewhat difficult to explain / operate for students

A-4
3 The Genetic Algorithm (GA)

A genetic algorithm mimics the process of evolution to create an optimum evolved solution to a
characterized problem. Criteria and encoded on chromosomal analogues that represent the
solution to the problem. Though a series of random mutations and subsequent evaluation
functions the grade the fitness of the chromosomes into achieving a near optimal solution.

The GA algorithm populates a single schedule randomly representing a single individual. This
individual is altered by random mutations, and its fitness is evaluated using a fitness function.
The modified individual is either kept or discarded. The final schedule for the courses is
displayed along with its fitness value.

Advantages:
• Optimized solution where both student and admin goals are met
• Admin gets good publicity in attempting to improve course selection system
• Easy to implement – Much open source software is available

Disadvantages:
• Somewhat computationally complex
• More difficult to explain / operate for students

A-5

Вам также может понравиться