stable marriage problem - florida institute of technologypkc/cs4hs/million/stablemarriage.pdf · 1/...
TRANSCRIPT
1/ 58
Stable Marriage ProblemA Million Dollar Problem
Ryan Stansifer
Department of Computer SciencesFlorida Institute of TechnologyMelbourne, Florida USA 32901
http://www.cs.fit.edu/~ryan/
25 March 2017
2/ 58
A matching algorithm
A ranking algorithm
A ???? problem
3/ 58
Problem 1:known as the
stable marriage problem
Solution: known as thedeferred acceptance algorithm
4/ 58
Shapely and Roth won the 2012 Nobel Prize in Economics
5/ 58
10 million SEKUS$ 1.4 millione 950,000
6/ 58
The problem is to find mutually acceptable matching of n things ofone kind to n things of another. For example:
donors to receipients,
residents to hospital opportunities,
students to available spots in schools, and
men to women
7/ 58
Let us meet the men and the women.
Let us make a matching.
8/ 58
AMY BETH CLARE DOT ELLA FRAN
ZACK
YORK
XAVIER
WALTER
VICTOR
ULRIC
9/ 58
AMY BETH CLARE DOT ELLA FRAN
ZACK ZACK CLARE
YORK
XAVIER
WALTER
VICTOR
ULRIC
10/ 58
AMY BETH CLARE DOT ELLA FRAN
ZACK ZACK CLARE
YORK YORK ELLA
XAVIER
WALTER
VICTOR
ULRIC
11/ 58
AMY BETH CLARE DOT ELLA FRAN
ZACK ZACK CLARE
YORK YORK ELLA
XAVIER XAVIERBETH
WALTER
VICTOR
ULRIC
12/ 58
AMY BETH CLARE DOT ELLA FRAN
ZACK ZACK CLARE
YORK YORK ELLA
XAVIER XAVIERBETH
WALTER
VICTOR VICTOR DOT
ULRIC
13/ 58
AMY BETH CLARE DOT ELLA FRAN
ZACK ZACK CLARE
YORK YORK ELLA
XAVIER XAVIERBETH
WALTER WALTERAMY
VICTOR VICTOR DOT
ULRIC
14/ 58
AMY BETH CLARE DOT ELLA FRAN
ZACK ZACK CLARE
YORK YORK ELLA
XAVIER XAVIERBETH
WALTER WALTERAMY
VICTOR VICTOR DOT
ULRIC ULRIC FRAN
15/ 58
There are many possible matching.
AMY BETH CLARE DOT ELLA FRAN
ZACK ZACK DOT
YORK YORK BETH
XAVIER XAVIER ELLA
WALTER WALTERFRAN
VICTOR VICTORCLARE
ULRIC ULRIC AMY
Several example matchings example 1: WXZVYU men emoji 16
AMY BETH CLARE DOT ELLA FRAN
ZACK ZACK BETH
YORK YORK ELLA
XAVIER XAVIERCLARE
WALTER WALTERFRAN
VICTOR VICTOR DOT
ULRIC ULRIC AMY
Several example matchings example 2: UZXVYW men emoji 17
AMY BETH CLARE DOT ELLA FRAN
ZACK ZACK ELLA
YORK YORKCLARE
XAVIER XAVIER AMY
WALTER WALTERFRAN
VICTOR VICTOR DOT
ULRIC ULRIC BETH
Several example matchings example 3: XUYVZW men emoji 18
AMY BETH CLARE DOT ELLA FRAN
ZACK ZACK AMY
YORK YORK BETH
XAVIER XAVIERCLARE
WALTER WALTERDOT
VICTOR VICTORELLA
ULRIC ULRIC FRAN
Several example matchings example 4: ZYXWVU men emoji 19
20/ 58
There is always a matching of all.
Is there a best one?
For that we need to consider the preferences of the all the men andwomen.
21/ 58
Men’s Preferences
1 2 3 4 5 6
ZACK CLARE AMY BETH ELLA FRAN DOT
YORK ELLA CLARE BETH AMY FRAN DOT
XAVIER BETH CLARE AMY ELLA FRAN DOT
WALTER AMY CLARE BETH ELLA FRAN DOT
VICTOR CLARE DOT FRAN BETH AMY ELLA
ULRIC AMY FRAN DOT CLARE BETH ELLA
22/ 58
Women’s Preferences
1 2 3 4 5 6
AMY ULRIC WALTER XAVIER YORK ZACK VICTOR
BETH ULRIC VICTOR XAVIER YORK ZACK WALTER
CLARE ZACK XAVIER YORK VICTOR ULRIC WALTER
DOT YORK XAVIER ZACK VICTOR ULRIC WALTER
ELLA VICTOR XAVIER YORK ZACK WALTER ULRIC
FRAN ZACK XAVIER YORK VICTOR ULRIC WALTER
23/ 58
We can look at a matchingas a table for the men or for the women.
24/ 58
AMY BETH CLARE DOT ELLA FRAN
ZACK ZACK BETH
YORK YORK ELLA
XAVIER XAVIERCLARE
WALTER WALTERFRAN
VICTOR VICTOR DOT
ULRIC ULRIC AMY
25/ 58
ZACK YORK XAVIER WALTER VICTOR ULRIC
AMY ULRIC AMY
BETH ZACK BETH
CLARE XAVIERCLARE
DOT VICTOR DOT
ELLA YORK ELLA
FRAN WALTERFRAN
26/ 58
Recall:We are looking for a good matching.
It not possible for everyone to get their first choice. But it wouldbe nice to eliminate dissatification leading to instability.
What do we mean? Consider the following example . . .
27/ 58
A B C D E F
CABEFD Zack BethECBAFD York EllaBCAEFD Xavier ClareACBEFD Walter FranCDFBAE Victor DotAFDCBE Ulric Amy
Z Y X W V U
UWXYZV Amy UlricUVXYZW Beth ZackZXYVUW Clare XavierYXZVUW Dot VictorVXYZWU Ella YorkZXYVUW Fran Walter
28/ 58
A B C D E F
CABEFD Zack BethECBAFD York EllaBCAEFD Xavier ♥ ClareACBEFD Walter FranCDFBAE Victor DotAFDCBE Ulric Amy
Z Y X W V U
UWXYZV Amy UlricUVXYZW Beth Zack ♥ZXYVUW Clare XavierYXZVUW Dot VictorVXYZWU Ella YorkZXYVUW Fran Walter
29/ 58
A B C D E F
CABEFD Zack Beth ♥ECBAFD York EllaBCAEFD Xavier ClareACBEFD Walter FranCDFBAE Victor DotAFDCBE Ulric Amy
Z Y X W V U
UWXYZV Amy UlricUVXYZW Beth ZackZXYVUW Clare ♥ XavierYXZVUW Dot VictorVXYZWU Ella YorkZXYVUW Fran Walter
30/ 58
Deferred Acceptance Algorithm
Is there an algorithm that will matchall the men and women efficiently with no unhappy couples?
31/ 58
Deferred Acceptance Algorithm
Repeat the following steps until all are matched.
An unmatched man proposes to the more preferred womanwho has not previously rejected his proposal.
Then, a woman (tentatively) accepts the proposal breaking anexisting engagement when the new suitor is preferred (if any).
A B C D E F
CABEFD ZackECBAFD YorkBCAEFD XavierACBEFD WalterCDFBAE VictorAFDCBE Ulric
Z Y X W V U
UWXYZV AmyUVXYZW BethZXYVUW ClareYXZVUW DotVXYZWU EllaZXYVUW Fran
Deferred acceptance algorithm step 0: no matches 32
A B C D E F
CABEFD Zack ClareECBAFD YorkBCAEFD XavierACBEFD WalterCDFBAE VictorAFDCBE Ulric
Z Y X W V U
UWXYZV AmyUVXYZW BethZXYVUW Clare ZackYXZVUW DotVXYZWU EllaZXYVUW Fran
Deferred acceptance algorithm step 1: Zack proposes to Clare 33
A B C D E F
CABEFD Zack ClareECBAFD York EllaBCAEFD XavierACBEFD WalterCDFBAE VictorAFDCBE Ulric
Z Y X W V U
UWXYZV AmyUVXYZW BethZXYVUW Clare ZackYXZVUW DotVXYZWU Ella YorkZXYVUW Fran
Deferred acceptance algorithm step 2: York proposes to Ella 34
A B C D E F
CABEFD Zack ClareECBAFD York EllaBCAEFD Xavier BethACBEFD WalterCDFBAE VictorAFDCBE Ulric
Z Y X W V U
UWXYZV AmyUVXYZW Beth XavierZXYVUW Clare ZackYXZVUW DotVXYZWU Ella YorkZXYVUW Fran
Deferred acceptance algorithm step 3: Xavior proposes to Beth 35
A B C D E F
CABEFD Zack ClareECBAFD York EllaBCAEFD Xavier BethACBEFD Walter AmyCDFBAE VictorAFDCBE Ulric
Z Y X W V U
UWXYZV Amy WalterUVXYZW Beth XavierZXYVUW Clare ZackYXZVUW DotVXYZWU Ella YorkZXYVUW Fran
Deferred acceptance algorithm step 4: Walter proposes to Amy 36
A B C D E F
CABEFD Zack ClareECBAFD York EllaBCAEFD Xavier BethACBEFD Walter AmyCDFBAE VictorAFDCBE Ulric
Z Y X W V U
UWXYZV Amy WalterUVXYZW Beth XavierZXYVUW Clare ZackYXZVUW DotVXYZWU Ella YorkZXYVUW Fran
Deferred acceptance algorithm step 5: Clare rejects Victor’s proposal 37
A B C D E F
CABEFD Zack ClareECBAFD York EllaBCAEFD Xavier BethACBEFD Walter AmyCDFBAE Victor DotAFDCBE Ulric
Z Y X W V U
UWXYZV Amy WalterUVXYZW Beth XavierZXYVUW Clare ZackYXZVUW Dot VictorVXYZWU Ella YorkZXYVUW Fran
Deferred acceptance algorithm step 6: Victor proposes to Dot 38
A B C D E F
CABEFD Zack ClareECBAFD York EllaBCAEFD Xavier BethACBEFD WalterCDFBAE Victor DotAFDCBE Ulric Amy
Z Y X W V U
UWXYZV Amy UlricUVXYZW Beth XavierZXYVUW Clare ZackYXZVUW Dot VictorVXYZWU Ella YorkZXYVUW Fran
Deferred acceptance algorithm step 7: Ulric proposes to Amy 39
A B C D E F
CABEFD Zack ClareECBAFD York EllaBCAEFD Xavier BethACBEFD WalterCDFBAE Victor DotAFDCBE Ulric Amy
Z Y X W V U
UWXYZV Amy UlricUVXYZW Beth XavierZXYVUW Clare ZackYXZVUW Dot VictorVXYZWU Ella YorkZXYVUW Fran
Deferred acceptance algorithm step 8: Clare rejects Walter’s proposal 40
A B C D E F
CABEFD Zack ClareECBAFD York EllaBCAEFD Xavier BethACBEFD WalterCDFBAE Victor DotAFDCBE Ulric Amy
Z Y X W V U
UWXYZV Amy UlricUVXYZW Beth XavierZXYVUW Clare ZackYXZVUW Dot VictorVXYZWU Ella YorkZXYVUW Fran
Deferred acceptance algorithm step 9: Beth rejects Walter’s proposal 41
A B C D E F
CABEFD Zack ClareECBAFD York EllaBCAEFD Xavier BethACBEFD WalterCDFBAE Victor DotAFDCBE Ulric Amy
Z Y X W V U
UWXYZV Amy UlricUVXYZW Beth XavierZXYVUW Clare ZackYXZVUW Dot VictorVXYZWU Ella YorkZXYVUW Fran
Deferred acceptance algorithm step 10: Ella rejects Walter’s proposal 42
A B C D E F
CABEFD Zack ClareECBAFD York EllaBCAEFD Xavier BethACBEFD Walter FranCDFBAE Victor DotAFDCBE Ulric Amy
Z Y X W V U
UWXYZV Amy UlricUVXYZW Beth XavierZXYVUW Clare ZackYXZVUW Dot VictorVXYZWU Ella YorkZXYVUW Fran Walter
Deferred acceptance algorithm step 11: Walter proposes to Fran 43
All are matched.No blocking pairs.
There might be many stable matchings(but in this particular case there are not)
A B C D E F
CABEFD Zack Clare •ECBAFD York EllaBCAEFD Xavier BethACBEFD Walter FranCDFBAE Victor DotAFDCBE Ulric Amy
Z Y X W V U
UWXYZV Amy UlricUVXYZW Beth XavierZXYVUW Clare ZackYXZVUW Dot • VictorVXYZWU Ella YorkZXYVUW Fran Walter
Blocking Pair? No. Dot-Zack 45
A B C D E F
CABEFD Zack ClareECBAFD York EllaBCAEFD Xavier BethACBEFD Walter • FranCDFBAE Victor DotAFDCBE Ulric Amy
Z Y X W V U
UWXYZV Amy • UlricUVXYZW Beth XavierZXYVUW Clare ZackYXZVUW Dot VictorVXYZWU Ella YorkZXYVUW Fran Walter
Blocking Pair? No. Amy-Walter 46
47/ 58
Definition
A pair (m,w) is blocking for a matching if m prefers w to hismatch and also w prefers m to her match.
A blocking pair destablizes a matching as the pair have incentiveto abandoned their partners.
48/ 58
Definition
A matching is perfect if all are paired up.
Definition
A matching is stable if there are no blocking pairs.
It is not obvious, but there is always a stable matching no matterwhat the preferences are.
Theorem
There is always a perfect, stable matching.
49/ 58
Definition
A man and a woman are possible for each other if some stablematching marries them.
Theorem (Gale-Shapley)
Every man gets his best possible match.
Theorem (Gale-Shapley)
Every woman gets her worse possible match.
50/ 58
Theorem (Dubins-Freedman)
No man or consortium of men can improve their results in themale-proposing algorithm by submitting false preferences.
51/ 58
Efficient Implmetnation
Does woman w prefer man m to man m′?Create inverse of preference list of men.
Amy 1 2 3 4 5 6
pref 6(U) 4(W) 3(X) 2(Y) 1(Z) 5(V)
Amy 1 2 3 4 5 6
inv 5 4 3 2 1 1
for i=1 to n
inv [pref[i]] = i
Amy prefers man 6 to man 3 since inv[6]=1 ¡ inv[3]=3
52/ 58
Social Utility
There are many ways to do matching. But the remarkable thingabout the deferred acceptance algorithm is that:
1 there is no advantage to lie or cheat about the preference, and
2 no swap will improve all the parties involved
This means making matchings this way can reduce strife and otherproblems.
53/ 58
National Resident Matching Program (NRMP)www.nrmp.org
NRMP places applicants for postgraduate medical trainingpositions into residency programs at teaching hospitals throughoutthe United States.
54/ 58
Historical Context
Hospitals made offers earlier and earlier and residents waitedlonger and longer to accept.
In 1952 a centralized clearing house (NRMP) was formed.
Originally used the “Boston Pool” algorithm. Mostlyequivalent to the deferred acceptance algorithm whichhospital optimality.
Algorithm overhauled and new algorithm adopted May 1997.
Resident optimalSide constraints (couples, unacceptability, unequal numbers)
Today over 42,000 residents for over 30,000 positions.
55/ 58
56/ 58
In New York City in the late 1990s tens of thousands of childrenwere placed in badly matched schools.
The process was so byzantine it appeared nothingshort of a Nobel Prize-worthy algorithm could fix it.Three economists Atila Abdulkadiroglu (Duke), ParagPathak (M.I.T.) and Alvin E. Roth (Stanford), all expertsin game theory and market design were invited to attackthe sorting problem together. Their solution was a modelof mathematical efficiency and elegance.
In 2014 75,000 eight grade students applied for a spot at one of426 public high schools, in a few hours the Class of 2019 wasarranged.
57/ 58
College Admissions
58/ 58
Kidney
59/ 58
National Kidney RegistryFacilitating Living Donor Transplantshttp://www.kidneyregistry.org/
The National Kidney Registry uses the power oftechnology and large pools of donor/recipient pairs tofind better matches through paired exchange.
60/ 58
The top trading cycle algorithm may produce long chains (SeeNYT) and the maximal matching algorithm can be distorted by ahospital’s goal of minimizing its costs over the greater social good.In general, the goal is to incentive hospitals to report all of theirpatient-donor pairs, to save as many lives as possible.
References I
D. Gale and Lloyd S. Shapley. “College Admissions and theStability of Marriage”. In: American Mathematical Monthly120.5 (May 2013), pp. 386–391. doi: http://dx.doi.org/10.4169/amer.math.monthly.120.05.386. url:http://www.jstor.org/stable/pdfplus/10.4169/amer.
math.monthly.120.05.386.pdf.
D. Gusfield and R. W. Irving. The Stable Marriage Problems:Structure and Algorithms. Cambridge, Massachusetts: MITPress, 1989.
Grace Lyo. The Stable Marriage Problem. Girls’ AngleWomen in Mathematics Video Series. Presented by EmilyRhiel. 2012. url: http://www.girlsangle.org/page/filmpage.php?num=16.
Numberphile. The Stable Marriage Problem. YouTube.Presented by Emily Rhiel. 2014. url:https://www.youtube.com/watch?v=Qcv1IqHWAzg.
61/ 58
References II
Alvin E. Roth. “The Economics of Matching: Stability andIncentives”. In: Mathematics of Operation Research 7.4(2009), pp. 79–112.
Alvin E. Roth. “The Evolution and the Labor Market forMedical Interns and Residents: A Case Study in GameTheory”. In: Political Economy Y2.6 (1984).
Alvin E. Roth. “The Evolution of the Labor Market forMedical Interns and Residents: A Case Study in GameTheory”. In: Journal of Polical Enconomy 92 (1984),pp. 991–1016. url: http://www.economics.harvard.edu/~aroth/papers/evolut.pdf.
Alvin E. Roth and Elliot Peranson. “The Redesign of theMatching Market for American Physicians: Some EngineeringAspects of Ecnomic Design”. In: The American EconomicReview 89.4 (1999).
62/ 58
References III
Alvin E. Roth, T. Sonmez, and M. U. Unver. “KidneyExchange”. In: Quaterly Journal of Economics 119 (2004),pp. 457–488.
Alvin E. Roth, T. Sonmez, and M. U. Unver. “PairwiseKidney Exchange”. In: Journal of Economic Theory 125(2005), pp. 151–1888.
Kevin Sack. “60 lives, 30 kidneys, all linked”. In: New YorkTimes (February 2012). url:http://www.nytimes.com/2012/02/19/health/lives-
forever-linked-through-kidney-transplant-chain-
124.html.
63/ 58
References IV
Tracey Tullis. “How Game Thoery Helped Improve New YorkCity’s High School Application Process”. In: New YorkTimes (December 2014). url:https://www.nytimes.com/2014/12/07/nyregion/how-
game-theory-helped-improve-new-york-city-high-
school-application-process.html.
64/ 58