Constraint Satisfaction Problems &
Constraint Programming
Representing Knowledge with Constraints
x1
x2
x6
x3
x5
KNOWLEDGE REPRESENTATION & REASONING
x4
1
KR&R for Combinatorial Problems


Many, many practical applications
 Resource allocation,
scheduling,
routing, frequency
 Sports
scheduling
example
assignment, timetabling,
vehicle
routing,
etc.
 In sports
league
scheduling
we try to build the schedule of
matches between teams (e.g. football teams). There are
Properties
various constraints:
 Computationally difficult
 Each team must play each other exactly twice (once home
once away)
 Technical and modeling and
expertise
needed
 Experimental in nature No team can play more than two consecutive home or away
matches
 Important ($$$) in practice
 The number of times that a team plays two consecutive
 Many solution techniques
home or away matches must be minimum
 Teams that use the same stadium cannot play home games
 (Mixed) integer programming
at the same date
 Specialized methods
 Games between top teams must occur at certain dates (due
 Local search/metaheuristics
to TV coverage)
 Constraint programming
 Etc.
KNOWLEDGE REPRESENTATION & REASONING
2
Quotations

“Constraint programming represents one of the closest
approaches computer science has yet made to the Holy
Grail of programming: the user states the problem, the
computer solves it.”
Eugene C. Freuder, Constraints, April 1997

“Were you to ask me which programming paradigm is
likely to gain most in commercial significance over the
next 5 years I’d have to pick Constraint Programming,
even though it’s perhaps currently one of the least
known and understood.”
Dick Pountain, BYTE, February 1995
KNOWLEDGE REPRESENTATION & REASONING
3
Constraint Programming

Constraint programming is the
continuous dream of programming
x1
State the constraints
 The solver will find a solution


Knowledge representation with
constraints is the preferred model in
many domains
Scheduling
 Vehicle routing
 Resource allocation
 Temporal Reasoning
 …

KNOWLEDGE REPRESENTATION & REASONING
x2
x6
x3
x5
x4
4
What is Constraint Programming?

Broad Answer:

Programming where the use of constraints plays a central role
 alternative to logic programming, functional programming, object-
oriented programming
 there are constraint programming languages that support this

What is a constraint?
Let X1,X2, . . . ,Xn be a finite sequence of variables, each associated with
a domain, D1,D2, . . . ,Dn.
 A constraint on X1,X2, . . . ,Xn is a relation D1 × D2 × · · · × Dn

 can be defined explicitly/extensionally or implicitly/intentionally
KNOWLEDGE REPRESENTATION & REASONING
5
What is Constraint Programming?

Narrow Answer:
A more specific answer is obtained by programming with
constraints in a particular manner.
 Constraint programming involves solving a problem by:

 Modelling: Formulate the problem as a finite set of constraints (a
Constraint Satisfaction Problem).
 Solving: Solve the CSP, perhaps by using a constraint programming
language
 Mapping: Map the solution to the CSP to a solution to the original
problem
KNOWLEDGE REPRESENTATION & REASONING
6
What is a Constraint?

What is a constraint?
Let X1,X2, . . . ,Xn be a finite sequence
of variables, each associated with a
domain, D1,D2, . . . ,Dn.
 A constraint on X1,X2, . . . ,Xn is a
relation D1 × D2 × · · · × Dn

 can be defined explicitly/extensionally
or implicitly/intentionally
There are constraints everywhere!
explicit
(0,1,0)
(1,0,0)
(1,1,0)
(1,1,1)
implicit
x+y>z





KNOWLEDGE REPRESENTATION & REASONING
Room Myrto is occupied from 12:00 until 15:00
Traffic in the web < 100 Gbytes/sec
Salary < 15k Euro
Train 1 must leave 20 minutes before train 2 arrives
Exams for 1st semester must be at least 2 days apart
7
Constraint Programming (CP)

Began in late 1970s – early 1980s
from AI world
III (Marseilles, France)
 CLP(R)
 CHIP (ECRC, Germany)
These days…
 Prolog

Software Engineering
AI
Application areas
CP
 Scheduling,
sequencing, resource
and personnel allocation, etc. etc.

Active research area
 Specialized
conferences (CP,
CP/AI-OR, …)
 Journal (Constraints)
 Companies (ILOG, COSYTEC,..)
KNOWLEDGE REPRESENTATION & REASONING
Logic Programming
OR
Discrete Mathematics
Let’s look at this
8
Constraint Programming

Two main contributions

A new solving approach to combinatorial problems
 AI search methods and heuristics
 Orthogonal and complementary to standard OR methods

A new language for representing combinatorial problems
 Rich language for constraints

Much closer to the real problem than OR
 Language for search procedures
 Easily extensible
KNOWLEDGE REPRESENTATION & REASONING
9
The Origins

Artificial Intelligence
 Scene
Operations Research
Interactive Graphics
Sketchpad (Sutherland)
 ThingLab (Borning)

Labelling (Waltz)




NP-hard combinatorial problems
Logic Programming

unification --> constraint solving
Let’s look at this
KNOWLEDGE REPRESENTATION & REASONING
10
Integer Programming (IP)
Consider the manufacture of television sets. A linear programming
model might give a production plan of 205.7 sets per week.


No trouble stating that production should be 205 sets per week (or even
``roughly 200 sets per week'').
Suppose we were buying warehouses to store finished goods. A model
that suggests we buy 0.7 warehouse at some location and 0.6 somewhere
else would be of little value.


Warehouses come in integer quantities, and we would like our model to reflect
that fact.
This integrality restriction has far reaching effects. Modeling with
integer variables has turned out to be useful far beyond restrictions to
integral production quantities. With integer variables, one can model
logical requirements, fixed costs, sequencing and scheduling
requirements, and many other problem aspects.

KNOWLEDGE REPRESENTATION & REASONING
11
Integer Programming (IP)

The trouble with all this modeling power, however, is that problems
with as few as 40 variables can be beyond the abilities of even the
most sophisticated computers.
Most real problems with more than 100 or so variables are not possible
to solve unless they show specific exploitable structure.
 Despite the possibility (or even likelihood) of enormous computing
times, there are methods that can be applied to solving integer programs.


An IP problem in which all variables are required to be integer is
called a pure integer programming problem.
If some variables are restricted to be integer and some are not then the problem
is a mixed integer programming problem (MIP).
 The case where the integer variables are restricted to be 0 or 1 is called pure
(mixed) 0-1 programming problems or pure (mixed) binary integer
programming problems.

KNOWLEDGE REPRESENTATION & REASONING
12
Relationship to Linear Programming
Given an integer program
There is an associated linear program called the linear relaxation formed by dropping the
integrality restrictions:
Since (LR) is less constrained than (IP), the following are immediate:
If (IP) is a minimization, the optimal objective value for (LR) is less than or equal to the
optimal objective for (IP).
 If (IP) is a maximization, the optimal objective value for (LR) is greater than or equal to that
of (IP).
 If (LR) is infeasible, then so is (IP).
Solving (LR) does give some information: it gives a bound on the optimal value, and, if we are
lucky, may give the optimal solution to IP. But for some problems it is very difficult to even
get a feasible solution!

KNOWLEDGE REPRESENTATION & REASONING
13
Branch and Bound
We will explain branch and bound by using this model:
Maximize 8X1+ 11X2 + 6X3 + 4X4
Subject to 5X1+ 7X2 + 4X3 + 3X4 ≤ 14
Xj {0,1} j = 1,…4.
The linear relaxation solution is X1=1, X2 = 1,X3 = 0.5, X4 =0 with a value of
22. We know that no integer solution will have value more than 22.
Unfortunately, since X3 is not integer, we do not have an integer solution yet.
We want to force X3 to be integer. To do so, we branch on X3, creating two
new problems. In one, we will add the constraint X3=0. In the other, we add
the constraint X3= 1.
Note that any optimal solution to the overall problem must be feasible to one
of the subproblems. If we solve the linear relaxations of the subproblems, we
get the following solutions:
X3 = 0: objective 21.65, X1 = 1, X2 = 1, X3 = 0 ,X4=0.677;
X3 = 1: objective 21.85, X1 = 1, X2 = 0.714, X3 = 1, X4 = 0.
At this point we know that the optimal integer solution is no more than
21.85, but we still do not have any feasible integer solution. So, we will take
a subproblem and branch on one of its variables. In general, we will choose
the subproblem as follows:
We will choose an active subproblem, which so far only means one we have
not chosen before, and we will choose the subproblem with the highest
solution value (for maximization) (lowest for minimization).
In this case, we will choose the subproblem with X3 = 1, and branch on X2.
After solving the resulting subproblems, we have the branch and bound tree
in Figure 2.
Figure 1
Figure 2
KNOWLEDGE REPRESENTATION & REASONING
14
Branch and Bound
The solutions are:
X3 = 1,X2 = 0: objective 18, X1 = 1, X2=0, X3=1,X4=1;
X3 = 1, X2=1: objective 21.8, X1 = 0.6, X2=1, X3=1, X4=0.
We now have a feasible integer solution with value 18.
Furthermore, since the X3=1, X2=0 problem gave an
integer solution, no further branching on that problem is
necessary. It is not active due to integrality of solution.
There are still active subproblems that might give values
more than 18. Using our rules, we will branch on problem
X3=1, X2=1 by branching on X1to get Figure 3.
The solutions are:
X3=1, X2=1, X1=0: objective 21, X1=0, X2=1, X3= 1, X4=1
X3 =1 , X2=1, X1=1: infeasible.
Our best integer solution now has value 21. The
subproblem that generates that is not active due to
integrality of solution. The other subproblem generated is
not active due to infeasibility. There is still a subproblem
that is active. It is the subproblem with solution value
21.65. There is no better integer solution for this
subproblem than 21. But we already have a solution with
value 21. It is not useful to search for another such
solution. Therefore, we can mark this subproblem it not
active. There are no longer any active subproblems, so the
optimal solution value is 21.
KNOWLEDGE REPRESENTATION & REASONING
Figure 3
15
Constraint Satisfaction
Problems
At the core of Constraint Programming
KNOWLEDGE REPRESENTATION & REASONING
16
What is a Constraint Satisfaction Problem?

A constraint satisfaction problem (CSP) is defined
by:


A set of variables X1,…,Xn
 Each variable Χi has a domain Di with its possible values
A set of constraints C1,…,Cm
 Each constraint involves a subset of the variables it specifies
the allowed combinations of values for this subset
1
 A k-ary constraint C on a set of variables X1,…,Xk is a subset of
the Cartesian product D1 x…x Dk
{0,…,5}
 The set of variables in a constraint is called the constraint
scope
x


x2
{0,…,3}
Binary and non-binary (or n-ary) constraint
satisfaction problems
KNOWLEDGE REPRESENTATION & REASONING
17
Constraint Satisfaction Problems

Solution of a CSP


Assignment of a value to each variable so that all constraints are satisfied
Goals:
Find one solution (feasibility problem)
 Find all solutions
 Find a solution that maximizes (or minimizes) some quantity

 constraint optimization problem

Find an approximate “solution”
All these tasks are NP-hard!
(except perhaps one of them)
KNOWLEDGE REPRESENTATION & REASONING
18
Constraint Graphs & Hypergraphs
x1
x2
x1
x2
x6
x3
x6
x4
x5


x5
x4
variables – nodes
binary constraints – edges
 the label of an edge specifies
the constraint
KNOWLEDGE REPRESENTATION & REASONING
x3


variables – nodes
n-ary constraints – hyperedges
19
Example – Map Coloring


We want to color each area in the map with a different color
We have three colors
red, green, blue
KNOWLEDGE REPRESENTATION & REASONING
20
Example – Map Coloring
Formal Definition:
 Variables



Domains (the same for all variables)


WA, NT, SA, Q, NSW, V, T
{red, green, blue}
Constraints
C(WA,NT) = {(red, green), (red, blue), (green, red), (green, blue),
(blue,red), (blue, green)}
 C(WA,SA) = …

KNOWLEDGE REPRESENTATION & REASONING
21
Constraint Graph
NT
Q
WA
SA
All constraints are
binary
NSW
Two unconnected
components
V
T
KNOWLEDGE REPRESENTATION & REASONING
22
Example – 8 Queens problem
We want to place 8
queens on the chessboard
so they can’t attack each
other

KNOWLEDGE REPRESENTATION & REASONING
23
Example– 8 Queens problem

Formal Definition:

Variables


Each variable Xi (i=1,…,8) represents the column where there is the i-th
queen (i.e. the queen in the i-th row)
Domains

If the columns are represented by numbers from 1 to 8 then the domain
of each variable Xi is
Di = {1,2,…,8}
KNOWLEDGE REPRESENTATION & REASONING
24
Παράδειγμα – 8 Queens problem

Constraints



There is a binary constraint C(Xi, Xj) for each pair of variables. These
constraints can be defined as follows:
For all variables Xi and Xj , Xi  Xj
For all variables Xi and Xj , if Xi = a and Xj = b then
i – j  a – b and i – j  b – a
KNOWLEDGE REPRESENTATION & REASONING
25
Example – Cryptoarithmetics
T WO
+T WO
FO U R
F
T
X3
KNOWLEDGE REPRESENTATION & REASONING
U
X2
W
R
O
X1
26
Example – Cryptoarithmetics
Formal Definition:
 Variables and Domains

F, T, U, W, R, O  {0,1,2,3,4,5,6,7,8,9}
 X1, X2, X3  {0,1}


Constraints
alldifferent(F, T, U, W, R, O)
 O + O = R + 10 X1
 X1 + W + W = U + 10 X2
 X2 + T + T = O + 10 X3
 X3 = F

KNOWLEDGE REPRESENTATION & REASONING
T WO
+T WO
FO U R
27
Example: Crossword puzzle
1
2
3
4
5
KNOWLEDGE REPRESENTATION & REASONING
28
Crossword puzzle as a CSP

Variables and their domains
X1 is 1 across
 X2 is 2 down
 X3 is 3 down
 X4 is 4 across
 X5 is 5 across


D1 consists of all 5-letter words in the dictionary
D2 consists of all 4-letter words in the dictionary
D3 consists of all 3-letter words in the dictionary
D4 consists of all 4-letter words in the dictionary
D5 consists of all 2-letter words in the dictionary
Constraints (implicit/intensional)
C12 is “the 3rd letter of X1 must equal the 1st letter of X2”
 C13 is “the 5th letter of X1 must equal the 1st letter of X3”
 C24 is …
 C25 is …
 C34 is ...

KNOWLEDGE REPRESENTATION & REASONING
29
Crossword puzzle as a CSP
1
2
3
4
Variables:
X1
X2
X3
X4
X5
5
Domains:
D1 = {astar, happy, hello, hoses}
D2 = {live, load, peal, peel, save, talk}
D3 = {ant, oak, old}
D4 = {live, load, peal, peel, save, talk}
KNOWLEDGE REPRESENTATION & REASONING
X1
X2
X3
X4
Constraints
(explicit/extensional):
C12 = {(astar, talk),
(happy, peal),
(happy, peel),
(hello, live) …}
C13 = ...
30
Real Constraint Satisfaction Problems

puzzles (not really practical applications, but they are fun)













N-queens, Zebra (five house puzzle), crossword puzzle, cryptoarithmetics
(SEND+MORE=MONEY), mastermind
graph coloring
analysis and synthesis of analog circuits
option trading analysis
cutting stock
DNA sequencing
crew scheduling
chemical hypothetical reasoning
warehouse location
patient treatment scheduling
airport counter allocation (Cathay Pacific Airways Ltd)
crew rostering problem (Italian Railway Company)
well activity scheduling (Saga Petroleum a.s.)
KNOWLEDGE REPRESENTATION & REASONING
31
Early Commercial Applications (90s)
Lufthansa: Short-term staff planning.
 Hongkong Container Harbor: Resource planning.
 Renault: Short-term production planning.
 Nokia: Software configuration for mobile phones.
 Airbus: Cabin layout.
 Siemens: Circuit verification.
 Caisse d’epargne: Portfolio management.

KNOWLEDGE REPRESENTATION & REASONING
32
Applications in Research

Artificial Intelligence

Machine Vision
 Natural Language Understanding
 Temporal and Spatial Reasoning
 Theorem Proving

 Qualitative Reasoning

 Robotics

 Agents

 Planning
Timetabling
Scheduling
Vehicle Routing
Resource allocation
 Frequency Assignment
KNOWLEDGE REPRESENTATION & REASONING
33
Applications in Research

Computer Science:


Molecular Biology, Biochemestry, Bioinformatics:


Parsing
Medicine:


Scheduling, Stock Investment Planning
Linguistics:


Protein Folding, Genomic Sequencing
Economics:


Program Analysis, Robotics, Agents
Decision Support
Physics: System Modeling
KNOWLEDGE REPRESENTATION & REASONING
34
CSP Technology : Practical & Successful

Constraint satisfaction technology is one of the most
successful examples of practical AI

There are many successful companies which build and trade
CSP technology
ILOG
 Cosytec
 Parc Technologies
 i2 Technologies
 IQ Software
 …

KNOWLEDGE REPRESENTATION & REASONING
35
CSP Technology : Practical & Successful!
















AKL
 FSQP/CFSQP
ALE
 Goedel
Amulet and Garnet
 GNU-Prolog
 ICE InC++ library
B-Prolog
 IF/Prolog
Bertrand
Brandeis Interval Arithmetic Constraint ILOG Numerica, ILOG Schedule, ILOG Solver
 Interval Solver for Microsoft Excel
Solver
 JSolver
 RISC-CLP(Real)
CHIP
 LIFE
 SEL
CIAL
 MAC
 ICStus
CLAIRE
 Newton
 Screamer
CLP
 Nicolog
 StarFLIP++
CONFLEX'
 Omega
 Steeles constraint system'
 Oz
CPLEX
 TOY
 ProFIT
 Toupie
Cassowary
 Prolog III, Prolog IV
 Trilogy
Contax
 Pulsar
 Unicalc
Cooldraw, Deltablue, Skyblue,
 QUAD-CLP(R)
 cu-Prolog
ThinglabII
 Quantum Leap
 opbdp
ECLiPSe
KNOWLEDGE REPRESENTATION & REASONING
36
A real CSP – Job-shop scheduling
Examples of job shop scheduling problems include
factory scheduling problems, in which some operations have to be
performed within one or several shifts
 spacecraft mission scheduling problems, in which time windows are
determined by astronomical events over which we have no control
 patient treatment scheduling problems, in which a number of patients
need to receive treatment that requires certain equipment within certain
time windows, etc.


When solving a job shop CSP, the objective is to find as
quickly as possible a feasible schedule, namely a schedule
where each operation is performed within one of its legal
time windows and no resource is oversubscribed.
KNOWLEDGE REPRESENTATION & REASONING
37
Job-shop scheduling problem (JSSP)

A JSSP requires scheduling a set of jobs J={ j1, ... , jn} on a
set of physical resources RES={R1, ... ,Rm}

Each job j consists of a set of operations O ={O1, ... ,On} to be scheduled
according to a process routing that specifies a partial ordering among
these operations (e.g. Oi BEFORE Oj ).
O1
O4
O6
O2
O5
O7
O3
O8
O1
O2
O3
Job 2
Job 1
KNOWLEDGE REPRESENTATION & REASONING
38
Job-shop scheduling problem (JSSP)



Each job j has a release date rdj and a due date (or deadline) ddj
between which all its operations have to be performed.
Each operation Oi has a fixed duration dui and a start time sti whose
value has to be selected.
The domain of possible start times of each operation is initially
constrained by the release and due dates of the job to which the
operation belongs.

there can be additional unary constraints that further restrict the set of
admissible start times of each operation, thereby defining one or several
time windows within which an operation has to be carried out
 e.g. a specific shift in factory scheduling

In order to be successfully executed, each operation Oi requires pi
different resources (e.g. a machine) Rij (1 j pi )
KNOWLEDGE REPRESENTATION & REASONING
39
The JSSP as a CSP

Variables

A set of variables is associated with each operation, Oi, which consists of
 the operation start time, sti
 its resource requirements, Rij

Constraints

Precedence constraints defined by the process routings translate into
linear inequalities of the type: sti +dui stj (i.e. Oi BEFORE Oj )
 Capacity constraints that restrict the use of each resource to only one
operation at a time translate into disjunctive constraints of the form: ("p
where Oi ,Oj require Rp) sti +dui stj stj +duj sti. These constraints
simply express that, unless they use different resources, two operations Oi
and Oj cannot overlap.
KNOWLEDGE REPRESENTATION & REASONING
40
The JSSP as a CSP
A job shop problem with 4 jobs
Each node is labeled by the operation
that it represents and the
resource required by this operation.
Each operation has a single resource
requirement with a single possible value.
Operation start times are the only variables.
KNOWLEDGE REPRESENTATION & REASONING
41
A real CSP – The car sequencing problem






In a car production scenario, cars are placed on conveyor belts which
move through different work areas.
A production line is normally required to produce cars of different
models. The number of cars required for each model is called the
production requirement.
Each work area is constrained by its resource constraint or Capacity
constraint.
variable – one for every position in the conveyor belt (i.e. if there are
n cars to be scheduled, the problem consists of n variables).
domain - the set of car models, for example from model A to D.
The task - to assign a value (a car model) to each variable (a position
in the conveyor belt), satisfying both the production requirements and
capacity constraints.
KNOWLEDGE REPRESENTATION & REASONING
42
The car sequencing problem
KNOWLEDGE REPRESENTATION & REASONING
43
Constraints and Databases

There are close links between CSPs and relational database theory
Constraint terminology Database terminology
CSP
Variable
Domain
Constraint
Constraint scope
Constraint tuples
Set of solutions
KNOWLEDGE REPRESENTATION & REASONING
Database
Attribute
Attribute domain
Table
Table schema
Table instance
Join of all tables
44
Constraints and Databases – Example

Consider the following CSP
A set of variables X = {x0,…,x9}
 All variables have the domain D = {0,1,2}
 There are constraints with the following scopes and allowed tuples:









c1 = {x0,x1,x3} – {(0,0,0), (0,1,0), (1,0,1), (1,1,1), (0,1,2)}
c2 = {x1,x2,x3} – {(0,0,0), (0,0,1), (1,1,0), (1,0,1), (0,1,2)}
c3 = {x1,x4} – {(0,0), (1,1)}
c4 = {x3,x6} – {(0,0), (1,1), (1,0), (2,0)}
c5 = {x4,x5,x6} – {(0,0,0), (0,0,1), (1,1,1), (1,0,2)}
c6 = {x4,x7} – {(0,1), (1,0)}
c7 = {x5,x8} – {(0,1), (1,0), (1,1)}
c8 = {x6,x9} – {(0,0), (1,1)}
KNOWLEDGE REPRESENTATION & REASONING
45
Constraints and Databases – Example
The constraints as a relational database
c1
c2
c3
c4
c5
c6
c7
c8
x0 x1 x3 x1 x2 x3 x1 x4 x3 x6 x4 x5 x 6 x4 x7 x5 x8 x6 x 9
0 0 0 0 0 0 0 0 0 0
0 1 0 0 0 1 1 1 1 0
1 0 1 1 1 0
1 1
1 1 1 1 0 1
2 0
e t
c.
0 1 2 0 1 2
KNOWLEDGE REPRESENTATION & REASONING
46
Solving CSPs

Assuming we have expressed knowledge about a problem as
a CSP
how can we reason with it?
 how can we find a solution (if one exists)?
 how can we find all solutions?
 how can we infer new knowledge?

Generate and test
 Backtracking search algorithms
 Approximation algorithms
 Constraint propagation algorithms

KNOWLEDGE REPRESENTATION & REASONING
47
Solving CSPs

There are two general approaches to solving CSPs
that are used in practice

Systematic Search
 Explore systematically the space of all assignments
 systematic = every valuation will be explored sometime
 extends partial assignments

Local Search
 explore the search space by small steps
 start with an initial complete assignment
 repairs complete assignments
KNOWLEDGE REPRESENTATION & REASONING
48
First of all: Generate & Test
Probably the most general problem solving method
 Algorithm:

generate labelling
test satisfaction
Drawbacks:
blind generator
late discovery of
inconsistencies
KNOWLEDGE REPRESENTATION & REASONING
Improvements:
smart generator
--> local search
testing within generator
--> backtracking
49
Generate and test: Crossword Puzzle

Try each possible combination until you find one that
works:
– live – ant – live
 astar – live – ant – load
 astar – live – ant – peal
…
 astar
1
2
3
4
5
Doesn’t check constraints until all variables have been
instantiated
 Very inefficient way to explore the space of possibilities
(4*6*3*6 = 432 for this trivial problem, most inconsistent)

KNOWLEDGE REPRESENTATION & REASONING
50
Search Algorithms for CSPs
A general search algorithm for CSPs
 Initial State



Actions


Assign a value from Di to an unassigned variable Xi
Goal Test


No value has been assigned to any variable
All variables have been assigned and all constraints are satisfied
The order in which the actions are executed does not matter

We can take advantage of this!
KNOWLEDGE REPRESENTATION & REASONING
51
The Search Space of CSPs
The search space is finite
 The depth of the search tree is specified



Equal to the number of variables
Solutions are always at the leaves of the search tree
Leaves
KNOWLEDGE REPRESENTATION & REASONING
52
Search Algorithms for CSPs

Which generic AI search algorithm looks suitable for CSPs?

Breadth-First Search ?


Depth-First Search ?


No! BFS will be inefficient because solutions are always at the leaves
Better than BFS. But it will frequently waste time searching while
constraints are already violated
Hill Climbing ?

Minimize conflicts
KNOWLEDGE REPRESENTATION & REASONING
53
Search Algorithms for CSPs

We will study variations of DFS especially for CSPs.

These algorithms are based on backtracking search
Simple (or Chronological) Backtracking (BT)
 Backjumping (BJ)
 Forward Checking (FC)
 FC with Conflict-based Backjumping (FC-CBJ)
 Maintaining Arc Consistency (MAC)

Also two variations of hill climbing


Min-conflicts
Min-conflicts with Random Walk
KNOWLEDGE REPRESENTATION & REASONING
54
Chronological Backtracking (ΒΤ)


The basic idea in all systematic backtracking-based algorithms is to
start with a partial solution (i.e. assignments of a subset of the
variables) and continue assigning variables until we reach a complete
solution
BT follows this technique
 Consider the variables in some order
 Pick an unassigned variable and give it a provisional value such
that it is consistent with all of the constraints
 If no such assignment can be made, we’ve reached a dead end and
need to backtrack to the previous variable and try its next value
 Continue this process until a solution is found or we backtrack to
the initial variable and have exhausted all possible valaues
KNOWLEDGE REPRESENTATION & REASONING
55
Chronological Backtracking (ΒΤ)
Previous variables
variable 0
{
variable 1
b
a
a
b
a
variable 2
b
(current variable)
{
b
a
b a
a
variable 3
b
a
b a
current assignment
b
a
variable 4
solution
b
Future variables
KNOWLEDGE REPRESENTATION & REASONING
56
Chronological Backtracking (ΒΤ)
procedure CHRONOLOGICAL_BACKTRACKING (vars,doms,cons)
solution  BT (vars,Ø,doms,cons)
function BT (unlabelled,compound_label,doms,cons)
returns a solution or NIL
if unlabelled = Ø then return compound_label
else pick a variable x from unlabelled
repeat
pick a value v from Dx; delete v from Dx
if compound_label + {(x,v)} violates no constraints then
result  BT(unlabelled - {x}, compound_label +
{(x,v)}, doms,cons)
if result  NIL then return result
end
until Dx = Ø
return NIL
end
KNOWLEDGE REPRESENTATION & REASONING
57
Chronological Backtracking (in action)
WA = red
WA = red
NT = red
WA = red
NT = green
Q = red
WA = red
NT = green
WA = red
NT = green
Q = green
WA = green
WA = blue
WA = red
NT = blue
WA = red
NT = green
Q = blue
KNOWLEDGE REPRESENTATION & REASONING
58
Backtracking: Crossword Puzzle
1
a s
4
2
t a
a
l
5
k
3
r
u
n

X1=astar
X1=happy
X2=live
X2=load
X3=ant
KNOWLEDGE REPRESENTATION & REASONING
…
X2=live
…
X2=talk
X3=oak
X3=old
59
Chronological Backtracking (ΒΤ)
Evaluation
 Complete and Sound ?



Time complexity: Ο(dne)


where d is the maximum domain size, n the number of variables, and e
the number of constraints
Χώρος: Ο(nd)


Yes and Yes
the space required to store the domains of all variables
The complexities hold under the assumption that all
constraint checks are performed in constant time and
constraints are stored in constant space
KNOWLEDGE REPRESENTATION & REASONING
60
GT & BT – Example 1

Problem:
X::{1,2}, Y::{1,2}, Z::{1,2}
X = Y, X  Z, Y > Z
generate & test
X
1
1
1
1
2
2
2
Y
1
1
2
2
1
1
2
Z
1
2
1
2
1
2
1
backtracking
test
fail
fail
fail
fail
fail
fail
passed
KNOWLEDGE REPRESENTATION & REASONING
X
1
Y
1
2
2
1
2
Z
1
2
1
test
fail
fail
fail
fail
passed
61
GT & BT 4-queen problem
Q1
1
2
3
4
Q2
Q3
Q4
Place 4 queens so that no two queens are
in attack.
Qi: line number of queen in column i, for 1i4
Q1, Q2, Q3, Q4 
Q1Q2, Q1Q3, Q1Q4,
Q2Q3, Q2Q4, Q3Q4,
Q1Q2-1, Q1Q2+1, Q1Q3-2, Q1Q3+2,
Q1Q4-3, Q1Q4+3,
Q2Q3-1, Q2Q3+1, Q2Q4-2, Q2Q4+2,
Q3Q4-1, Q3Q4+1
KNOWLEDGE REPRESENTATION & REASONING
62
4-queen problem first solution
Q1
Q2
Q3
Q4
1
2
3
4
There is a total of 256 valuations
GT algorithm will generate
64 valuations with Q1=1;
+
+
=
48 valuations with Q1=2, 1Q23;
3 valuations with Q1=2, Q2=4, Q3=1;
115 valuations to find first solution
KNOWLEDGE REPRESENTATION & REASONING
63
4-queen problem, BT algorithm
Q1
1
2
3
4
Q2
Q3
Q4
Q1
Q2
Q3
Q4
Q1
Q2
Q3
Q4
Q1
Q2
Q3
Q4
1
2
3
4
1
2
3
4
1
2
3
4
KNOWLEDGE REPRESENTATION & REASONING
Q1
Q2
Q3
Q4
1
2
3
4
64
Advantages

declarative nature with procedural capabilities
(when needed)


co-operative problem solving


unified framework for integration of variety of special-purpose
algorithms
semantic foundation



focus on describing the problem to be solved, and choosing the
algorithm to solve it
amazingly clean and elegant languages
roots in logic programming
applications

proven success
KNOWLEDGE REPRESENTATION & REASONING
65
Limitations
NP-hard problems & tractability
 unpredictable behaviour
 ad-hoc modelling
 too much expertise required


new constraints, solvers, heuristics, modelling
non-incremental (rescheduling)
 awkward handling of optimization



solvers tuned to finding first solution
weak solver collaboration

with OR engines for example
KNOWLEDGE REPRESENTATION & REASONING
66
Useful Links

On-line guide to Constraint Programming
 http://kti.ms.mff.cuni.cz/%7Ebartak/constraints/

Constraints Archive
 http://www.cs.unh.edu/ccc/archive/

CSPLib : a problem library for constraints
 http://4c.ucc.ie/~tw/csplib/

Course on Theory and Practice of
Constraint Satisfaction

http://www.cse.unl.edu/~choueiry/CSCE990-05/schedule.htm
KNOWLEDGE REPRESENTATION & REASONING
67
Descargar

Dynamic and Distributed Scheduling in Communication