drools cylande chtijug 2010

Copyright © Ch'ti JUG – License Creative Commons 2.0 France

Ch’ti JUGCh’ti JUG

Jboss Drools&

Drools Planner

21 janvier 2010

Ch’ti JUGCh’ti JUG ● Editeur de logiciels exclusivement dédiés aux enseignes du Retail

● Création en 1986

● 35 M€ de CA en 2008 (+20 %/an en moyenne depuis 5 ans)

● 34 % à l’International

● 5 sites en France dont le siège à Roubaix. (Paris, Belfort,

Antibes, Vannes)

● 5 filiales hors hexagone : Shanghai, Portugal, Espagne, Tunisie, Pologne en cours

● Une expérience éprouvée dans 60 pays


44

Effectifs : 430 collaborateurs dans le monde, 360 en France, 300 ressources basées à Roubaix

0

10

20

30

40

50

60

70

80

90

100

Effectifs

Experts Metiers

Directeur de Projet

Chef de Projet Métier

Chef de ProjetTechniques

Formateur et @learning

Directeurs de Produits

Développeurs

Recette et Qualification

Hot Line

Préparateurs etdéploiements

Ch’ti JUGCh’ti JUG Storeland pilote l’ensemble de votre supply chain étendue


Cylande a accompagné l’équipe de France

de Judo à Pékin

Le judo véhicule des valeurs

CYLANDE partenaire de la FFJ


Lauréat du Prix PME France CHINE ACFCI / CCIFC

Une croissance résolument tournée vers l’international

●The SkyNet funding bill is passed. ●The system goes online on August 4th, 1997.●Human decisions are removed from strategic defense. ●SkyNet begins to learn at a geometric rate.●It becomes self-aware at 2:14am Eastern time, August 29th ●In a panic, they try to pull the plug. ●And, Skynet fights back

Mark Proctor

Project Lead

9

Business Logic integration Platform

DroolsGuvnor

DroolsFusion

DroolsFlow

DroolsExpert

Introduction

Drools Expert

Learn by Example

12

D a t e d a t ed o u b l e a m o u n ti n t t y p el o n g a c c o u n t N o

C a s h f l o w

l o n g a c c o u n t N od o u b l e b a l a n c e

A c c o u n t

D a t e s t a r tD a t e e n d

A c c o u n t i n g P e r i o d

Classes

13

AccountaccountNo balance

1 0

increase balance for AccountPeriod Credits

select * from Account acc, Cashflow cf, AccountPeriod apwhere acc.accountNo == cf.accountNo and cf.type == CREDIT cf.date >= ap.start and cf.date <= ap.end

decrease balance for AccountPeriod Debits

select * from Account acc, Cashflow cf, AccountPeriod apwhere acc.accountNo == cf.accountNo and cf.type == DEBIT cf.date >= ap.start and cf.date <= ap.end

AccountingPeriodstart end

01-Jan-07 31-Mar-07

trigger : acc.balance += cf.amount

trigger : acc.balance -= cf.amount

Accountbalance

1 -25accountNo

Creating Views with Triggersdate amount type

12-Jan-07 100 CREDIT 12-Feb-07 200 DEBIT 118-May-07 50 CREDIT 19-Mar-07 75 CREDIT 1

accountNo

date amount type12-Jan-07 100 CREDIT9-Mar-07 75 CREDIT

CashFlow

date amount type2-Feb-07 200 DEBIT

CashFlow

14

What is a Rule

• rule “<name>” <attribute> <value> when <LHS> then <RHS>end

Quotes on Rule names are optional if the rule name has no spaces.

salience <int>agenda-group <string>no-loop <boolean>auto-focus <boolean>duration <long>....

RHS can be any valid java. Or MVEL. Other languages could be added.

15

Imperative vs Declarative

• public void helloMark(Person person) { if ( person.getName().equals( “mark” ) { System.out.println( “Hello Mark” ); }}

• rule “Hello Mark” when Person( name == “mark” ) then System.out.println( “Hello Mark” );end

LHS

RHS

specific passing of instances

Methods that must be called directly

Rules can never be called directly

Specific instances cannot be passed.

16

S h o w e r( t e m p e r a t u r e = = “ h o t ” )

P a t t e r n

F i e l d C o n s t r a i n t

R e s t r i c t i o n

E v a l u a t o rV a l u e

O b j e c t T y p e

F i e l d N a m e

What is a Pattern

17

rule “increase balance for AccountPeriod Credits” when ap : AccountPeriod() acc : Account( $accountNo : accountNo ) CashFlow( type == CREDIT, accountNo == $accountNo, date >= ap.start && <= ap.end, $ammount : ammount ) then acc.balance += $amount; end

select * from Account acc, Cashflow cf, AccountPeriod apwhere acc.accountNo == cf.accountNo and cf.type == CREDIT cf.date >= ap.start and cf.date <= ap.end

trigger : acc.balance += cf.amount

Pattern

Pattern Binding

field Binding

Literal Restriction

Variable Restriction

Multri Restriction - Variable Restriction

field Binding

Consequence (RHS)

Bringing it Together

18

AccountaccountNo balance

1 0

rule “increase balance for AccountPeriod Credits” when ap : AccountPeriod() acc : Account( $accountNo : accountNo ) CashFlow( type == CREDIT, accountNo == $accountNo, date >= ap.start && <= ap.end, $ammount : ammount ) then acc.balance += $amount; end

AccountingPeriodstart end

01-Jan-07 31-Mar-07

rule “decrease balance for AccountPeriod Debits” when ap : AccountPeriod() acc : Account( $accountNo : accountNo ) CashFlow( type == DEBIT, accountNo == $accountNo, date >= ap.start && <= ap.end, $ammount : ammount ) then acc.balance -= $amount; end

Rules as a “ view”

Accountbalance

1 -25accountNo

date amount type12-Jan-07 100 CREDIT 12-Feb-07 200 DEBIT 118-May-07 50 CREDIT 19-Mar-07 75 CREDIT 1

accountNo

date amount type12-Jan-07 100 CREDIT9-Mar-07 75 CREDIT

CashFlowdate amount type

2-Feb-07 200 DEBIT

CashFlow

19

Patterns in more details

CashFlow( type == “credit” )

$ap : AccountPeriod()

CashFlow( date >= $ap.start )


CashFlow( date >= $ap.start && <= $ap.end )


CashFlow( type == “credit”,

date >= $ap.start && <= $ap.end )

20

More Pattern Examples

Person( $age : age )

Person( age == ( $age + 1 ) )

Person( $age : age )

Person( eval( age == $age + 1 ) )

Person( $age1 : age )

Person( $age2 : age )

eval( $age2 == $age1 + 1 )

21

Person(age > 30 && < 40 || hair == “black”)

Person(pets[’rover’].type == “dog”)

Person(pets[0].type == “dog”)

Person(age > 30 && < 40 || hair in (“black”, “brown”) )

Person(pets contain $rover )

Person( (age > 30 && < 40 && hair == “black”)

||

(age > 50 && hair == “grey”) )

More Pattern Examples

22

What is a Production Rule System

ProductionMemory

WorkingMemory

Inference Engine

Pattern Matcher

Agenda(rules) (facts)

insertupdateretract

Repository of inserted Java instances

Codification of the business knowledge

Rules can change

on the fly

23

A c c o u n tA c c o u n t i n g P e r i o dC a s h f l o w

v i e w1 v i e w2

m a i n v i e w

T a b l e s

V i e w s

V i e w

A c c o u n tA c c o u n t i n g P e r i o dC a s h f l o w

r u l e1 r u l e2

a g e n d a

O b j e c t T y p e s

R u l e s

a g e n d a

Production Rule SystemApproximated by SQL and Views

24

rule “Print blance for AccountPeriod” salience -50 when ap : AccountPeriod() acc : Account( ) then System.out.println( acc.accountNo + “ : “ acc.balance ); end

Salience

Agenda1 increase balance

arbitrary2 decrease balance3 increase balance4 print balance

Conflict Resolution with Salience

25

rule “increase balance for AccountPeriod Credits” ruleflow-group “calculation” when ap : AccountPeriod() acc : Account( $accountNo : accountNo ) CashFlow( type == CREDIT, accountNo == $accountNo, date >= ap.start && <= ap.end, $ammount : ammount ) then acc.balance += $amount; end

rule “Print blance for AccountPeriod” ruleflow-group “report” when ap : AccountPeriod() acc : Account( ) then System.out.println( acc.accountNo + “ : “ acc.balance ); end

ruleflow-group

RuleFlow

26

Two Phase System

Working Memory Action

retract

modifyinsert

Agenda Evaluation

Select Rule to Fire

exit

No RuleFound

Fire Rule

Determine possible rules to

fire

RuleFound

Conditional Elements

28

not Bus( color = “red” )

From CE for Expressions

exists Bus( color = “red” )

forall ( $bus : Bus( floors == 2 )

Bus( this == $bus, color == “red” ) )

forall ( $bus : Bus( color == “red” ) )

From CEfor Expressions

30


rule “Find all the pets for a given owner”when $owner : Person( name == “mark” ) Pet( name == “rover” ) from $owner.pets

Using 'from' to reason over the nested list

31

'from' can work on any expression, not just a nested field on a bound variable.


rule “Find People for given zip code”when $zipCode : ZipCode() Person( ) from $hbn.getNamedQuery(“Find People”) .setParameters( [ “zipCode” : $zipCode ] ) .list()Hibernate session

Collect CE

33

Collect CE

rule "accumulate"

when

$list : List( intValue > 100 )

from collect( Bus( color == "red" ) )

then

print "red buses “ + $list;

end

Accumulate CE

35

Accumulate CErule "accumulate"

when

$sum : Number( intValue > 100 )

from accumulate( Bus( color == "red", $t : takings )

init( sum = 0 ),

action( sum += $t ),

result( sum ) )

then

print "sum is “ + $sum;

end

36

Accumulate CErule "accumulate"

when


from accumulate( Bus( color == "red", $t : takings ) sum( $t ) )

then


end

37

Accumulate CE

Patterns and CE's can be chained with 'from'

rule "collect"

when

$zipCode : ZipCode()


from accumulate( Bus( color == "red", $t : takings )

from $hbn.getNamedQuery(“Find Buses” )

.setParameters( [ “zipCode” : $zipCode ] )

.list(),

sum( $t ) )

then


end

TimersCalendars

39

Timers

rule “name” timer 1m30swhen $l : Light( status == “on” )then SendEmail( “turn the light off” )

rule “name” timer (int: 0 1m30)when $l : Light( status == “on” )then SendEmail( “turn the light off” )

40

Timers

rule “name” timer ( cron: 0 0/15 * * * * )when $l : Light( status == “on” )then sendEmail( “turn the light off” )

Field Name Mandatory? Allowed Values Allowed Special CharactersSeconds YES 0-59 , - * /Minutes YES 0-59 , - * /Hours YES 0-23 , - * /Day of month YES 1-31 , - * ? / L WMonth YES 1-12 or JAN-DEC , - * /Day of week YES 1-7 or SUN-SAT , - * ? / L #Year NO empty, 1970-2099 , - * /

41

Calendarsrule "weekdays are high priority" calendars "weekday" timer (int:0 1h)when Alarm()then send( "priority high - we have an alarm” );end

rule "weekend are low priority" calendars "weekend" timer (int:0 4h)when Alarm()then send( "priority low - we have an alarm” );end

Truth MaintenanceInference

43

TMS and Inferencerule "Issue Child Bus Pass"

when

$p : Person( age < 16 )

then

insert(new ChildBusPass( $p ) );

end

rule "Issue Adult Bus Pass"

when

$p : Person( age >= 16 )

then

insert(new AdultBusPass( $p ) );

end

Couples the logic

What happens when the Child stops being 16?

44

TMS and Inference Bad

● Monolithic● Leaky● Brittle integrity - manual maintenance

45

TMS and Inference A rule “logically” inserts an object When the rule is no longer true, the object is retracted.

when

$p : Person( age < 16 )

then

logicalInsert( new IsChild( $p ) )

end

when


then

logicalInsert( new IsAdult( $p ) )

end

de-couples the logic

Maintains the truth by automatically retracting

46


when

$p : Person( )

IsChild( person =$p )

then

logicalInsert(new ChildBusPass( $p ) );

end

rule "Issue Adult Bus Pass"

when


IsAdult( person =$p )

then

logicalInsert(new AdultBusPass( $p ) );

end

The truth maintenance cascades

47


when

$p : Person( )

not( ChildBusPass( person == $p ) )

then

requestChildBusPass( $p );

End

The truth maintenance cascades

48

TMS and Inference Good

● De-couple knowledge responsibilities● Encapsulate knowledge● Provide semantic abstractions for those encapsulation● Integrity robustness – truth maintenance

Tooling

50

Guided Editor

51

Interactive Debugging

52

Decision Tables

53

DSLs

54

DSLs

55

Rule Flow

Drools Fusion

57

$c : Custumer( type == “VIP” )BuyOrderEvent( customer == $c )

session.insert( event ) ;

Rule engines do not scale for CEP. They have a single point of insertion and are single threaded, CEP has concurrent streams of events.

Single Point of entry

Patterns, evaluate facts sequentially in

a single thread.

Scalability

58

$c : Custumer( type == “VIP )BuyOrderEvent( customer == $c ) from entry-point “Home Broker Stream”

Scalability

EntryPoint entryPoint = session.getEntryPoint( “Home Broker Stream” );entryPoint.insert( event ) ;

So lets allow multiple named entry points for those streams

So now we can insert different

streams concurrently

Patterns can now optional specify their

entry-point.

When not specified uses the “default”

entry-point

59

All Fact life-cycles must be managed by the user, so retractions are manual.

declare StockTick @role( event )end

declare StockTick @role( event ) @timestamp( timestampAttr )

companySymbol : String stockPrice : double timestampAttr : longend

Automatic Life-Cycle Management

Just use the declare statement to declare a type as an event

and it will be retracted when it is no longer needed

The declare statement can also specify an internal model, that

external objects/xml/csv map on to. We support Smooks and

JAXB

60

$c : Custumer( type == “VIP )$oe : BuyOrderEvent( customer == $c ) from entry-point “Home Broker Stream” BuyAckEvent( relatedEvent == $oe.id, this after[1s, 10s] $oe ) from entry-point “Stock Trader Stream”

● coincides

● before

● after

● meets

● metby

● overlaps

● overlappedby

● during

● includes

● starts

● startedby

● finishes

● finishedby

Operators

Rule engines do not have rich enough set of temporal comparison operators BackAckEvent must occur

between 1s and 10s 'after' BuyOrderEvent

The Full set of Operators are supported

61

Operators

62

$c : Custumer( type == “VIP )$oe : BuyOrderEvent( customer == $c ) from entry-point “Home Broker Stream” not BuyAckEvent( relatedEvent == $oe.id, this after[1s, 10s] $oe ) from entry-point “Stock Trader Stream”

Operators

Existing Drools 'not' Conditional Elements can be used to detect

non-occurrence of events

63

Rule engines react to events happening now, there is no temporal understanding of changes over time.

That isn't much without the ability to deal with aggregations, rules engines suck.

Sliding time windows

StockTicker( symbol == “RHAT” ) over window:time( 5s )

StockTicker( symbol == “RHAT” ) over window:length( 1000 )

5s

1000 tickers

64

Aggregations

Rule Engines do not deal with aggregations

$n : Number( intValue > 100 ) from accumulate( $s : StockTicker( symbol == “RHAT” ) over window:time( 5s ), average( $s.price ) )

Over 5 seconds

Aggregate ticker price for RHAT over last 5

seconds

The pattern 'Number' reasons 'from' the accumulate result

$n : accumulate( $s : StockTicker( symbol == “RHAT” ) over window:time( 5s ), average( $s.price ) > 100 )

We can use some sugar to reduce verbosity

Drools Flow

66

Drools Flow

Integration● From loose coupling (decision services)● To advance integration (process rules)

Unification● Rules and processes are different types of business

knowledge assets● Infrastructure

● Timers/Schedulers● Testing● Communication/Services

● Tooling ● IDE● repository, management● Auditing●

A workflow engine combining processes and rules

Truth MaintenanceInference

68

Rules and processes

loosely coupledtightly coupled

spec

ific

gene

ric

DecisionServices

ProcessRules

SC

OP

E

COUPLING

?

69

Business Logic Lifecycle

70

Example

71

RuleFlowGroup

Workflow can control my rules?

72

RuleFlowGroup

Rule Flow Group

73

RuleFlowGroup

74

Constraints

Java code constraint

75

Constraints

Rules can control my workflow?

LHS “when”Rule Constraint

76

Example

Business decisions are externalized using a decision service

rule Decision1 when // conditions then // actionsend



77

Example

What if there is a lot of business logic like this?






















78

RulesEngine

Flow of Control

ProcessEngine

79

RulesEngine

Inversion of Control

ProcessEngine

Age

nda

80

Self monitoring and adaptivedeclare ProcessStartedEvent

@role( event )

end

rule "Number of process instances above threshold" when

Number( nbProcesses : intValue > 1000 )

from accumulate(

e: ProcessStartedEvent( processInstance.processId == "com.sample.order.OrderProcess" )

over window:size(1h),

count(e) )

then

System.err.println( "WARNING: Nb of order processes in the last hour > 1000: " + nbProcesses );

end

81

Domain Specific Processes

82

Domain Specific Processes

83

Integrated debug and audit

Unified API

DefinitionsRuntime

Language

85

jBPMFile file = new File (“.....”); // file to XML process definition

ProcessDefinition processDefinition = ProcessDefinition.parseXmlString( IoUtils.FileToString( file ) );

ProcessInstance processInstance = new ProcessInstance(processDefinition);JessRete engine = new Rete();

FileReader file = new FileReader("myfile.clp");

Jesp parser = new Jesp(file, engine);

parser.parse(false);Esper

EPServiceProvider epService = EPServiceProviderManager.getDefaultProvider();

EPStatement countStmt = admin.createEPL( "...." );

countStmt.start();

Definitions

86

Drools Flow

KnowledegBuilder kbuilder = KnowledgeBuilderFactory.newKnowledgeBulider();

kbuilder.addResource( ResourceFactory.newClassPathResource( “myflow.drf”, ResourceType.DRF );

If ( kbuilder.hasErrors() ) { log.error( kbuilder.hasErrors().toString() );}

KnowledgeBase kbase = KnowledgeBaseFactory.newKnowledgeBase();kbase.addKnowledgePackages( kbase.getKnowledgePackages() );

Definitions

87

Drools Expert

KnowledegBuilder kbuilder = KnowledgeBuilderFactory.newKnowledgeBulider();

kbuilder.addResource( ResourceFactory.newClassPathResource( “myrules.drl”, ResourceType.DRL );

If ( kbuilder.hasErrors() ) { log.error( kbuilder.hasErrors().toString() );}

KnowledgeBase kbase = KnowledgeBaseFactory.newKnowledgeBase();kbase.addKnowledgePackages( kbase.getKnowledgePackages() );

Definitions

88

Drools Integration Deployment Descriptors

<change-set> <add> <resource source='classpath:myapp/data/myflow.drf' type='DRF' /> <resource source='http:myapp/data/myrules.drl' type='DRL' />

<resource source='classpath:data/IntegrationExampleTest.xls' type="DTABLE">

<decisiontable-conf input-type="XLS" worksheet-name="Tables_2" />

</resource> <add></change-set>

KnowledegBuilder kbuilder = KnowledgeBuilderFactory.newKnowledgeBulider();kbuilder.addResource( ResourceFactory.newFileResource( “changeset.xml”, ResourceType.ChangeSet );

Unified Event Model

Mixins Interface Markers

92

Stateful KnowledgeSession

93

Knowledge Runtime

94

Coming in 5.1 BPMN2 (80% of Workflow) Integration

● OSGi ready● Spring● Camel

Seamless Remoting Simulation/Testing New Rete Algorithm “true modify”

95

Spring + camel

from("direct:test-with-session").to("drools:sm/ksession1?dataFormat=drools-xstream");

96

Questions?Questions?• Dave Bowman: All right, HAL; I'll

go in through the emergency airlock.

• HAL: Without your space helmet, Dave, you're going to find that rather difficult.

• Dave Bowman: HAL, I won't argue with you anymore! Open the doors!

• HAL: Dave, this conversation can serve no purpose anymore. Goodbye. Joshua: Greetings, Professor

Falken.Stephen Falken: Hello, Joshua.Joshua: A strange game. The only winning move is not to play. How about a nice game of chess?

http://www.imdb.com/name/nm0001158/











Q&A

9



Automated planningwith Drools Planner

Geoffrey De SmetDrools Planner lead

"Do more with less."


Ch’ti JUGCh’ti JUG Agenda

Use cases of automated planning• N queens• Bin packaging• Employee shift rostering• Examination timetabling

Find the best solution• With Drools Planner

Calculate the score of a solution• With Drools


Ch’ti JUGCh’ti JUG N Queens: use case

Place n queens on a n-sized chess board

No 2 queens can attack each other


Ch’ti JUGCh’ti JUG N queens: partially solved

Score -1 for every 2 queens that can attack each other

Score = -2


Ch’ti JUGCh’ti JUG N queens: an optimal solution

Score = 0


Ch’ti JUGCh’ti JUG N queens: demo

Not optimized!• Hello world

example

Not a realplanning problem• I can make an

optimal solutionfor any n queenswithout a computer

• See Wikipedia


Ch’ti JUGCh’ti JUG NP complete

Yellow item goes in first (or last)• Why?• Not the largest size• Not the largest side• So why?

NP complete• A given solution can be verified

fast• No efficient way to find a solution

• Is there even a solution?


Ch’ti JUGCh’ti JUG Real world bin packaging

Not just 5 items• 1000+ items

Not just 1 container• 100+ containers• Different container types

More constraints...• Distribute weight evenly• Not all fireworks in the same container• ...


Ch’ti JUGCh’ti JUG Hard constraints

Hard constraints must be fulfilled For example:

• Ensure continuous service• At least 1 emergency nurse at any given time

• Labor laws• Every 24 hours: at least 11 hours rest• Every 7 days: at least 35 hours rest

• No shifts during approved vacation


Ch’ti JUGCh’ti JUG Soft constraints

Soft constraints should be fulfilled as much as possible• Only after the hard constraints are

fulfilled

Each soft constraint is weighted For example:

• Fair night work assignment: weight 5• Forward rotation: weight 10• Nurse preferences: weight 1

• Ann dislikes Saturday night shifts• Beth dislikes Wednesday afternoon shifts


Ch’ti JUGCh’ti JUG Hard and soft score

Solution A B C Hard constraints

• 11 hours rest 1 0 0 Soft constraints

• Fair night workassignment 0 1000 1• Weight 5

• Nurse preferences 0 0 4000• Weight 1

Total score -1H/0S 0H/-5000S 0H/-4005S

• A < B < C• C is the best solution


Ch’ti JUGCh’ti JUG Hard constraints

Exam conflict: 2 exams that share students should not occur in the same period.

Room capacity: A room's seating capacity should suffice at all times.

Period duration: A period's duration should suffice for all of its exams.

Period related hard constraints should be fulfilled:

• Coincidence: 2 exams should use the same period (but possibly another room).

• Exclusion: 2 exams should not use the same period.

• After: 1 exam should occur in a period after another exam's period.

Room related hard constraints should be fulfilled:

• Exclusive: 1 exam should not have to share its room with any other exam.


Ch’ti JUGCh’ti JUG Soft constraints

2 exams in a row.

2 exams in a day.

Period spread: 2 exams that share studentsshould be a number of periods apart.

Mixed durations: 2 exams that share a roomshould not have different durations.

Front load: Large exams should be scheduledearlier in the schedule.

Period penalty: Some periods have a penalty when used.

Room penalty: Some rooms have a penalty when used.


Ch’ti JUGCh’ti JUG Examination demo

International timetabling competition 2007• Finished 4th (back then)

7 minutes• CPU depended

Real word test data 14 constraints

• 7 hard constraints• 7 soft constraints


Ch’ti JUGCh’ti JUG Other use cases

Vehicle routing• Freight routing

Scheduling• Course, meeting, conference scheduling• Appointment and resource scheduling• Sport scheduling

Storage organizing Machine queue planning ...


Ch’ti JUGCh’ti JUG Why use Drools Planner?

Open source• ASL (business-friendly)

Maven-ready (JBoss repository) Documentation

• Reference manual• Examples

JBoss Drools community support• User mailing list, issue tracking, …• Blog & twitter (#droolsplanner)


Ch’ti JUGCh’ti JUGBrute force

for (periodOfExam1 : periodList) { exam1.setPeriod(periodOfExam1); for (roomOfExam1 : roomList) { exam1.setRoom(roomOfExam1);

for (periodOfExam2 : periodList) { exam2.setPeriod(periodOfExam2); for (roomOfExam2 : roomList) { exam2.setRoom(roomOfExam2); ... for (periodOfExamN : periodList) { examN.setPeriod(periodOfExamN); for (roomOfExamN : roomList) { examN.setRoom(roomOfExamN);

Score score = calculateScore(solution); cloneIfScoreIsBetter(solution, score);

} } } }…} }


Ch’ti JUGCh’ti JUG Needle in a haystack

How many possible solutions?• 1096 exams• 80 periods• 28 rooms

> habitants in Lille per km²?• 6 483 hab./km²Source: wikipedia




> humans?• 7.000.000.000 humans

Source: NASA (wikipedia)




> minimum atoms in the observable universe?• 10^80 atoms

Source: NASA and ESA (wikipedia)




> atoms in the universeif every atom is a universe of atoms?• (10^80)^80 = 10^6400

Source: NASA and ESA (wikipedia)


Ch’ti JUGCh’ti JUG Do the math

1 exam• 80 periods and 28 rooms• 80 * 28 = 2240 ways to schedule 1 exam

2 exams• 2240 * 2240 = 5.017.600

3 exams• 2240 * 2240 * 2240 = 11.239.424.000

1096 exams• 2240 * 2240 * … * 2240• 2240^1096 = a little over 10^3671


Ch’ti JUGCh’ti JUGA little over

10^3671

74443724674464882011383315953154621497427697455114051316288269134692843108344990310502102147434076562448130852404428098553211787226818492436455899991484967631419697684165817985739661390634926254859096857258977301840109249945418286726701389433250396830489437134122748296147216955996361597777271017137683780046154870127217758740223489170130893779085381647394360334935333289368078384002213161233225755719910067066354676237665251240673552315376749902467736827879981604429943150088424040897721698276067946148250230917492054728443158872165054373936157659332956136774730870081258025518405492389480888615900164269035398348299000380567467552410280857265893710574057117390411923324486282853392817922617168734507604739703552080299261320457186755798353796720329958815466662988845983738466048902038122152381226870228697167564520947170314014038670253281783219898668392349799158354071694433128608374231159613003286648446078922185727592075724811



10^3671

6048135772412471854625105630495358121952017974176215221261550607694499282872000580072957918546796819172012885232741311107156500439895658139217642528073069419950416303276042981944782604076520149545429082567515199635531168668927010363569188258631683061394017239747010858770816458215631819437872729831119114113689168267734458648249288525981253268712682909721892541332433788104618254995718184937280503163787574781545179918774455713682720486085676323080374894817073654077307783490409626446740500738118392110173307114879831341215304834099815901166729699407017252645417836852601401021510814954906747082633216854492531462935276329826288243709434523924561625262847747165433198090950514642269855008208195099600705166755800356942782663732953126879621138033542807009649872210605061596144967082523007946872878429586274134471258439206305573503782097081716925686154420223798946020972887359043006100852387795351482973307623581925846555002793841



10^3671

412819475399046707554915331636124476210270759983783881007403725028189106738399600287059413396296063538199837169373556801830583664641156130483672354172652266198330743819868438588044621805009480956563538464893798379308830824383808936545111608312964868056598674131595193654957707706822143338172833633019666638035983430262037019665125647894212392790462389810030266845803079031515302062019379538886948677023472435462645765005804746816166402399340231002187005109182016211164762492991719240503935116392473986075551679379460553477047460526845933176425584932086637889540004159744719173226633548555732700361980207696413126618655189183160162357390484834785168386038147341617149224158994590819150108545695234158875676738936645877760000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000



10^3671

00000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000



10^3671

0000000000000000000000000000000000000000000000000000000000000000000000000000

The search space is big!• Compare with WWW size

• 22 020 000 000 pages

Each possible solution• 1096+ exams scheduled into

• 80 periods• 28 rooms

• Still need to calculate the score


Ch’ti JUGCh’ti JUG Throw hardware at it?

If 10^9 scores calculated per ms• Not possible today!• 31.579.200.000 ms in 1 year

• < 10^11 ms in 1 year

• 10^9 * 10^11 scores per year• = 10^20 scores per year

How many years? 10^3671 / 10^20• = 10^3651 years

CPU 1000 times faster• It becomes 10^3648 years


Ch’ti JUGCh’ti JUG A dose of reality

Find the optimal solution?• Of a real world planning problem?

Not in our lifetimes! Who cares?

• Beat the human planner(s) (=easy)• Spend less resources

• Save more money• Save the environment

• Make more people happy

• Never ending competition


Ch’ti JUGCh’ti JUG Smarter brute force?

Eliminate subtrees• Branch and bound• Still too many for loops• Still takes forever

for (periodOfExam2 : periodList) { exam2.setPeriod(periodOfExam2); if (exam1.shareStudentWith(exam2) && periodOfExam1.equals(periodOfExam2)) { continue; // bug: best solution might break a hard constraint } ...


Ch’ti JUGCh’ti JUGImperfect algorithms

(mimic a human)

Deterministic• First in, first assigned, never changed• Easy to implement

• Drools Planner score support

• Fixed time (for example 18 seconds)

Metaheuristic• Move things around

• Start from result of deterministic algorithm

• Drools Planner implementations• More time = better score


Ch’ti JUGCh’ti JUG Deterministic: N queens

Demo Not feasible

• Not optimal

Good initialization• Jump 10 meter into the

pool


Ch’ti JUGCh’ti JUG Deterministic: examination

List<Exam> sortedExamList = sortExamsOnDifficulty(examList);for (exam : sortedExamList) { // Determine best remaining spot Score bestScoreOfExam = - INFINITY; for (period : periodList) { exam.setPeriod(period); for (room : roomList) { exam.setRoom(room);

Score score = calculateScore(solution); if (score > bestScoreOfExam) { bestScoreOfExam = score; ... store bestPeriod, bestRoom } } } … assign exam to bestPeriod, bestRoom}


Ch’ti JUGCh’ti JUG Metaheuristic algorithms

Local search: 1st , 2nd , 3rd and 4th in ITC 2007• Simple local search (Hill climbing)• Tabu search

• Local search ++

• Simulated annealing• Great deluge• ...

Genetic algorithms: 5th in ITC 2007 Ant colony optimization ...


Ch’ti JUGCh’ti JUG Move things around

Move = from solution A to solution B• Change the row of 1 queen

• Give 2 queens each others rows• ...


Ch’ti JUGCh’ti JUG All moves from one solution

Number of moves < number of solutions• N queens

• n*n < n^n

• 4 queens• 16 < 256

• 8 queens• 64 < 16777216

• 64 queens• 4096 < 10^116


Ch’ti JUGCh’ti JUG Metaheuristic: local search


Ch’ti JUGCh’ti JUG Local optima

1) Deterministic StartingSolutionInitializer

2) Simple local search 3) Stuck in local optimum!

Source: Wikipedia


Ch’ti JUGCh’ti JUG Tabu search = local search++

Solution tabu (high tabu size)• Been there, no need to go there again

Move tabu (low tabu size)• Done that recently, no need to do that

again

Property tabu (low tabu size)• Changed that recently,

no need to change that again


Ch’ti JUGCh’ti JUG Drool planner configuration

<selector> <selector> <moveFactoryClass>...PeriodChangeMoveFactory</...> <relativeSelection>0.002</relativeSelection> </selector> ... <selector> <moveFactoryClass>...ExamSwitchMoveFactory</...> <relativeSelection>0.002</relativeSelection> </selector> </selector> <accepter> <completeSolutionTabuSize>1000</completeSolutionTabuSize> <completeMoveTabuSize>7</completeMoveTabuSize> </accepter> <forager> <foragerType>MAX_SCORE_OF_ALL</foragerType> </forager>


Ch’ti JUGCh’ti JUG Termination

Synchronous (configured)• Max timeMillis/seconds/minutes/hours

spend• Score attained• Max step count• Max unimproved step count

Asynchronous (from another thread)• planner.terminateEarly();


Ch’ti JUGCh’ti JUG Double time !=> double score

Softscore

Time (hours:minutes)

Examination test data 7


Ch’ti JUGCh’ti JUG Benchmarker utility

Battle of different planner configurations• Different algorithms (tabu search, ...)• Different moves• Different settings

On multiple datasets Results are ranked:

• Best one wins

Coming soon:• Graph: best score over time


Ch’ti JUGCh’ti JUG JAVA vs SQL vs DRL

for (q1 : queenList) { for (q2 : queenList) { if (q1.getId() < q2.getId() && q1.getY() == q2.getY()) { ... } }}

select *from Queen q1, Queen q2where q1.id < q2.id and q1.y = q2.y;

rule "multipleQueensHorizontal" when $q1 : Queen($id : id, $y : y); $q2 : Queen(id > $id, y == $y);


Ch’ti JUGCh’ti JUG N queens: score rule

rule "multipleQueensHorizontal" when $q1 : Queen($id : id, $y : y); $q2 : Queen(id > $id, y == $y); then insertLogical(new IntConstraintOccurrence( "multipleQueensHorizontal", ConstraintType.NEGATIVE_HARD, 1, $q1, $q2));end


Ch’ti JUGCh’ti JUG Score rule isolation

rule "multipleQueensHorizontal" when $q1 : Queen($id : id, $y : y); $q2 : Queen(id > $id, y == $y); then ...endrule "multipleQueensAscendingDiagonal" when $q1 : Queen($id : id, $ascendingD : ascendingD); $q2 : Queen(id > $id, ascendingD == $ascendingD); then ...endrule "multipleQueensDescendingDiagonal" when $q1 : Queen($id : id, $descendingD : descendingD); $q2 : Queen(id > $id, descendingD == $descendingD); then ...end


Ch’ti JUGCh’ti JUG Examination: period spread

2 exams that share students should be a number of periods apart

rule "periodSpread" when $iw : InstitutionalWeighting(periodSpreadPenality != 0); // For any 2 conflicting exams in the same period ... $topicConflict : TopicConflict($leftT : leftTopic, $rightT : rightTopic); $leftExam : Exam(topic == $leftT, $leftPeriod : period); $rightExam : Exam(topic == $rightT, $rightPeriod : period); // … which are in within the periodSpread eval(Math.abs($leftPeriod.getPeriodIndex() - $rightPeriod.getPeriodIndex()) < ($iw.getPeriodSpreadLength() + 1)); then insertLogical(new IntConstraintOccurrence(... NEGATIVE_SOFT, $topicConflict.getStudentSize() * $iw.getPeriodSpreadPenality(), $leftExam, $rightExam));end


Ch’ti JUGCh’ti JUG Summary

Drools Planner solves planning problems

Adding constraints is easy and scalable

Switching/combining algorithms is easy



Q&A

66

Home page• http://www.jboss.org/drools/drools-planner.html

Reference manual• http://www.jboss.org/drools/documentation.html

Blog• http://blog.athico.com/search/label/planner

Twitter• #droolsplanner

http://www.jboss.org/drools/drools-planner.html

http://www.jboss.org/drools/documentation.html

http://blog.athico.com/search/label/planner



Thanks for your attention!

67

Home page• http://www.jboss.org/drools/drools-planner.html

Reference manual• http://www.jboss.org/drools/documentation.html

Blog• http://blog.athico.com/search/label/planner

Twitter• #droolsplanner

http://www.jboss.org/drools/drools-planner.html

http://www.jboss.org/drools/documentation.html

http://blog.athico.com/search/label/planner


Ch’ti JUGCh’ti JUG Licence

Les photos et logos appartiennent à leurs auteurs respectifs

Le contenu de la présentation est sous licence Creative Commons 2.0 France• Contrat Paternité• Pas d'Utilisation Commerciale• Partage des Conditions Initiales à

l'Identique http://creativecommons.org/licenses/by-nc-sa/2.0/fr/

68


Ch’ti JUGCh’ti JUG Cocktail

Merci pour votre attention Merci à Cylande pour son sponsoring

69

drools cylande chtijug 2010

Technology