1 Course Scheduling Software Progress Presentation December 22, 2004.
-
Upload
marianna-fisher -
Category
Documents
-
view
212 -
download
0
Transcript of 1 Course Scheduling Software Progress Presentation December 22, 2004.
1
Course Scheduling SoftwareProgress Presentation
December 22, 2004
2
Customer:
Jed Lippard, Upper School Director,Prospect Hill Academy Charter School
Team [email protected]
Glen Winston Robert McKeever
Steve MoranValdeva Ives
3
Agenda• Project Status• Risk Update• Architectural Overview• GUI Walkthrough• Model Overview• Scheduling
– Proof of Concept– Constraint Programming– Technology Alternatives– Drools pros & cons– Drools Example– Scheduling API
• Deployment Plan
4
Project Status
Presented by
Glen Winston
5
Project Status
6
Process Overview• Project Phases: Fall
– Initiation (100% complete)– Analysis (100% complete)– Functional Design (100% complete)– Technical Design (70% complete)
• Project Phases: Spring– Completion of Technical Design– Development– Testing– Deployment
7
Process Overview: Technical Design• Key Deliverable of this phase
– Technical Specifications Document• Document all non-trivial classes in the system.• Informative sequence diagrams of key controllers
in the system.• Document and prove algorithm for scheduling
classes.
8
Risk Update
Presented by
Steve Moran
9
Risks• Scheduling technology implementation – higher
– Advancing Rule Base during semester break– Continuing Research into Production Systems & Constraints– Assigning 50% of team to task
• Scheduling technology choice – moderate– Several well understood alternatives available
• Post-Deployment Maintenance & Support – low to moderate– Specification biased toward currently stated requirements– Will rely on actual implementation
• Added Requirement to schedule individual students – lower– Students primarily move in groups – Believe less involved than class scheduling
• Feature creep – lower– Detailed documentation & customer contact
• Insufficient time - lower– Following regimented process
10
Component Overview
Presented by
Glen Winston
11
Architectural Overview
View Controller
Model
Scheduler
javax.swingjavax.swing
Scheduler API
State ChangeState Query
Change Notification
View Selection
User Gestures
Update State
12
GUI Walkthrough
Presented by
Dee Dee Ives
13
Model Overview
Presented by
Glen Winston
14
Model Overview
View Controller
Model
Scheduler
javax.swingjavax.swing
Scheduler API
State ChangeState Query
Change Notification
View Selection
User Gestures
Update State
15
Subject
name : String
Course
name : Stringsubject : SubjectblocksPerWeekdoubleBlocksPerWeekdepartments : List
Instructor
departments : ListmandatoryAvailability : WeeklyScheduledesiredAvailability : WeeklySchedulemaximumDailyClasses : int
Block DoubleBlock
Department
name : String
WeeklySchedule
days : ListperiodsPerDay
Day
name : Stringperiods : List
Period
day : DaytimePeriod
name : String
16
Cohort
students : Listgrade : Grade
Student
firstName : StringlastName : Stringgrade : Grade
SchoolClass
course : Courseinstructors : Listcohorts : Liststudents : List
Grade
name : String
Course
name : Stringsubject : SubjectblocksPerWeekdoubleBlocksPerWeekdepartments : List
17
Room
building : Stringfloor : Stringnumber : Stringcapacity : intavailability : WeeklySchedule
Block DoubleBlock
Period
day : DaytimePeriod
name : String
SchoolSchedule
name : StringclassSchedule : ClassSchedule
saveToFile()exportTeacherView()exportStudentView()importStudents()
ClassPeriod
schoolClasses : Listroom : Roomperiod : Periodfixed : booleanscheduled : boolean
ClassSchedule
classPeriods : List
findAvailableRooms(filter)findPeriods(filter)findClassPeriods(filter)
Created by Scheduling Algorithm
18
Scheduling Proof of Concept: Project Goal
Presented by
Glen Winston
19
Scheduling Proof of Concept: Project Goal
Scheduler Component is highest area of risk in technical design.
Risk Mitigation Plan
• Proof of concept• Two exclusive resources• Hand written algorithm fallback plan
Proof of Concept is complete, we learned
• We were able to create periods in a rules engine.
• We were able to fill periods in a rules engine.
20
Scheduling Proof of Concept: Approach
Presented by
Steve Moran
21
Scheduling Problem To Solve• Students, Teachers, & Subjects,• in 7 grades, subdivided into cohorts (groups),• into classrooms of various sizes & locations,• with 7 daily schedule blocks,• with a rotating class schedule,• with 81 initial constraints.
22
Technology Alternatives• DROOLS – actively being prototyped
– “understandable” XML style rules in java
• JESS – capable, but licensing issues • CLIPS – “lisp style” rules implemented in C• JClips – directly runs CLIPS files in java
– testbed for existing CLIPS.clp example files
• Prolog – inefficient backward chaining• Brute force - inefficient backward chaining
23
Rete-based Inference Engine• Declarative programming – “what is”• Forward chaining rules – “data driven”
• Fast in-memory network – “The only algorithm for implementing production rules
whose performance is demonstrably independent of the number of rules.”
• Rules can change without recompiling
24
<parameter>
Domain
<condition>
Boolean
<consequence>
java
Rules Working MemoryFacts
Assert
Retract
Modify
Facts Rules
Ordered by Salience
Rete
AlgorithmCollections of Objects
Classes
Rooms
Blocks
In a nutshell, we want to schedule Classes into Rooms with Blocks
25
RoomsRooms
Blocks
222
101 - 8AM
101
222
101
101 - 9AM101 - 10AM
222 - 8AM222 - 9AM
222 - 10AM
Before Rule Fires After Rule Fires
<parameter> Room </parameter>
<condition> none </condition>
<consequence>
assert(new Block(room.num,8))
assert(new Block(room.num, 9))
assert(new Block(room.num, 10))
</consequence>
Working
Memory
26
room 101 has block at 8
room 101 has block at 9
room 101 has block at 10
room 222 has block at 8
room 222 has block at 9
room 222 has block at 10
27
Rooms ClassBlocks
222101 - 8AM
101
Math
English101 - 9AM
101 - 10AM
222 - 8AM222 - 9AM
222 - 10AM
Working Memory
<parameter>
Room
Block
Class
</parameter>
<condition>
block.class = null
class.numStudents < room.capacity
class.isScheduled == false
</condition>
<consequence>
class.isScheduled = true
modify(class)
block.class = class.id
modify(block)
</consequence>
28
Scheduling class: 5 in room: 101 at: 8
Scheduling class: 15 in room: 101 at: 9
Scheduling class: 6 in room: 101 at: 10
Scheduling class: 8 in room: 222 at: 8
Scheduling class: 16 in room: 222 at: 9
Scheduling class: 20 in room: 222 at: 10
29
Scheduling Proof of Concept: Challenges
Presented by
Bob McKeever
30
Constraint Programming Problem• Scheduling is an NP Complete Problem.• Requires polynomial time to solve. • Could be solved trivially by using a systemic
search. • Generate and test until a solution is found.
31
Constraint Programming Solutions • Backtracking.• Backtracking with Forward Checking. • Backtracking with Forward Checking and
Heuristics. • Tic, Tac, Toe as an example
32
Drools Negatives • Very little documentation.• Does not have all the same features as
CLIPS. (At present “not” is not supported.)• Can not directly convert from CLIPS code to
Drools code. • Small user community.• Just out of Beta.
33
Drools Negatives (Con’t)• Team members have no experience with
Drools programming.• 3 Team members have no experience with
programming expert systems.
34
Drools Positives • Handles the constraints well. Much better
than nested if statements. • Open source.• Seems to have a lot of “buzz”.• Did I mention it was free?
35
Drools Positives (Con’t) • We are starting to get up to speed with it.
Now have some working examples.• May be able to post our work as an example
on their web site to have others carry on. May help on maintenance issues.
36
Scheduling Proof of Concept: Drools Sample
Presented by
Bob McKeever
37
<?xml version="1.0" encoding="UTF-8"?>
<!--
The definition of a RuleExecutionSet is not within the scope of the JSR 94.
The implementation given in this file is written for the reference
implementation. A rule engine vendor verifying their rule engine should
modify this file to their specific needs.
-->
<rule-set name="Scheduler"
xmlns="http://drools.org/rules"
xmlns:java="http://drools.org/semantics/java"
xmlns:xs="http://www.w3.org/2001/XMLSchema-instance"
xs:schemaLocation="http://drools.org/rules rules.xsd
http://drools.org/semantics/java java.xsd">
<java:import>java.util.*</java:import>
<java:import>org.drools.examples.schedule.model.Block</java:import>
<java:import>org.drools.examples.schedule.model.ClassInfo</java:import>
<java:import>org.drools.examples.schedule.model.ClassToSchedule</java:import>
<java:import>org.drools.examples.schedule.model.Room</java:import>
<java:import>org.drools.examples.schedule.model.RoomCourseRelation</java:import>
<java:import>org.drools.examples.schedule.model.RoomInfo</java:import>
<java:import>org.drools.examples.schedule.model.SchoolClass</java:import>
38
<!--
Create the blocks
-->
<rule name="generate blocks" salience="40">
<parameter identifier="roomInfo">
<class>RoomInfo</class>
</parameter>
<java:consequence>
System.out.println("Making block " + roomInfo.number);
drools.assertObject(new Block(roomInfo.number, 8));
drools.assertObject(new Block(roomInfo.number, 9));
drools.assertObject(new Block(roomInfo.number, 10));
</java:consequence>
</rule>
39
<!--
Schedule.
-->
<rule name="schedule" >
<parameter identifier="roomInfo"> <class>RoomInfo</class> </parameter>
<parameter identifier="block"> <class>Block</class> </parameter>
<parameter identifier="classInfo"> <class>ClassInfo</class> </parameter>
<java:condition> block.schoolClass == 0 </java:condition>
<java:condition> classInfo.numStudents <= roomInfo.capacity </java:condition>
<java:condition> classInfo.isScheduled == false </java:condition>
<java:consequence>
classInfo.isScheduled = true;
drools.modifyObject(classInfo);
block.schoolClass = classInfo.id;
drools.modifyObject(block);
System.out.println("Scheduling class: " + block.schoolClass +
" in room: " + block.room + " at: " + block.time);
</java:consequence>
</rule>
</rule-set>
40
Scheduling API
Presented by
Bob McKeever
41
Scheduling API
Scheduler
View Controller
Model
javax.swingjavax.swing
Scheduler APIScheduler API
State ChangeState Query
Change Notification
View Selection
User Gestures
Update State
42
43
Deployment Plan
Presented by
Bob McKeever
44
Deployment Plan Goals
• Keep customer informed. • Get buy in from customer’s IT administrator. • Create Windows executable.• Provide physical program to the customer.• Provide documentation to the customer. • Give the program the Windows look and feel.
45
Deployment Plan Actions• Use launch4j to create a Windows executable
with Splash screens and icons.• Meet with customer’s IT administrator.
Request computer that has been backed up. • Create set up program using Wise for install
and uninstall. Burn onto a CD.• Test!, Test!, Test!
46
Deployment Plan Actions (Con’t)• Develop documentation and detailed
installation instructions.• Provide professional documentation and CD.• Meet at Customers Site for installation. • Provide a Jar file version on the Web site of a
generic scheduler. (One that does not have the customer’s logos on it and is not dependent on Windows to run.)
47
Q & A
48
Backward & Forward Chaining
Presented by
Steve Moran
49
Two Approaches
• Backward Chaining– Imperative based systems – how to– Queries fact space for goal ‘truth’– Mechanism used in most most logic programming, i.e. Prolog
• Forward Chaining– Declarative based systems – what is– Triggered on fact space information– A data driven technique to reach inferences from a set of facts
50
Backward vs. Forward Chaining• Backward-chaining means that no rules are fired upon assertion of new
knowledge. When an unknown piece of knowledge is detected all rules relevant to the knowledge in question are fired until the question is answered, if possible. Thus, backward chaining systems normally work from a goal state back to the original state.
• Forward-chaining implies that upon assertion of new knowledge, all relevant rules are fired exhaustively, effectively making all knowledge about the current state explicit within the state. Forward chaining may be regarded as progress from a known state (the original knowledge) towards a goal state(s).
• The branching factor (the number of considerations at each state) may vary between forward and backward chaining and thus determine which method is most efficient.
Source: http://ai.eecs.umich.edu/cogarch0/index.html
51Source: http://www.cise.ufl.edu/class/cap6685sp03/Lectures/ES-CH4b.pdf
PatientDiagnosis
52
Backward vs. Forward Chaining• Backward-chaining means that no rules are fired upon assertion of new
knowledge. When an unknown piece of knowledge is detected all rules relevant to the knowledge in question are fired until the question is answered, if possible. Thus, backward chaining systems normally work from a goal state back to the original state.
• Forward-chaining implies that upon assertion of new knowledge, all relevant rules are fired exhaustively, effectively making all knowledge about the current state explicit within the state. Forward chaining may be regarded as progress from a known state (the original knowledge) towards a goal state(s).
• The branching factor (the number of considerations at each state) may vary between forward and backward chaining and thus determine which method is most efficient.
Source: http://ai.eecs.umich.edu/cogarch0/index.html
53Source: http://www.cise.ufl.edu/class/cap6685sp03/Lectures/ES-CH4b.pdf
54Source: http://www.cise.ufl.edu/class/cap6685sp03/Lectures/ES-CH4b.pdf
55Source: http://www.cise.ufl.edu/class/cap6685sp03/Lectures/ES-CH4b.pdf
56
Backward vs. Forward Chaining• Backward-chaining means that no rules are fired upon assertion of new
knowledge. When an unknown piece of knowledge is detected all rules relevant to the knowledge in question are fired until the question is answered, if possible. Thus, backward chaining systems normally work from a goal state back to the original state.
• Forward-chaining implies that upon assertion of new knowledge, all relevant rules are fired exhaustively, effectively making all knowledge about the current state explicit within the state. Forward chaining may be regarded as progress from a known state (the original knowledge) towards a goal state(s).
• The branching factor (the number of considerations at each state) may vary between forward and backward chaining and thus determine which method is most efficient.
Source: http://ai.eecs.umich.edu/cogarch0/index.html