Op 27-05-11 22:39, Michael Anstis schreef:
Phew! I made it to the end of the email ;)
All very interest Geoffrey, but have you explained the purpose of the
annotations? Why must I now annotate things, what value is it bringing?
The real value will be explained in the manual later and visible when we
build stuff on this :)
This mail is mainly to rattle people's cages:
*can you live with the bad side of these changes or not?*
Speak up now, before I go further on this path and there's no going back.
I don't, personally, have any problem with them - there are many
libraries that use them - but their purpose is well understood: e.g.
@Entity, @Resource etc
Good point that I need to explain it better to open up the discussion.
Basically, Planner needs to know what are you planning variables (of the
planning entity classes) and what are allowed values of each planning
variable (ignoring the non-build-in constraints).
It's also the first step to declarative moves.
Perhaps I missed something?
With kind regards
Mike
On 27 May 2011 11:13, Geoffrey De Smet <ge0ffrey.s...@gmail.com
<mailto:ge0ffrey.s...@gmail.com>> wrote:
Hi guys
For Drools Planner 5.3.0 (not the upcoming release)
I am working on a separate branch [1] to allow Planner to
understand your domain model better.
*This will force some big changes upon you as a user*, but it will
also allow many new features.
Reading and replying to this mail is your chance to steer those
changes, and verify that they are a good thing for your
implementation too.
I'll push those changes to master in a day or 2 ... unless someone
finds a good reason not too.
Any feed-back, especially on concept names, is welcome.
*The Bad News*
You'd have to do some serious upgrading changes. Although I am
confident this can be done in an hour or 2.
Here is the upgrading recipe as in github. Please go through to
this list to understand the impact of these changes.
Once the changes are on master, I 'll update the reference manual.
[MAJOR] You need to define your solution class in the
configuration now:
Before in *SolverConfig.xml and *BenchmarkConfig.xml:
<localSearchSolver>
<scoreDrl>...</scoreDrl>
After in *SolverConfig.xml and *BenchmarkConfig.xml:
<localSearchSolver>
<solutionClass>org.drools.planner.examples.curriculumcourse.domain.CurriculumCourseSchedule</solutionClass>
<scoreDrl>...</scoreDrl>
[RECOMMENDED] Understand the concept of a "planning entity" class.
The class (or classes) that change during planning (and do not
implement Solution) are a planning entity.
For example: ShiftAssignment, BedDesignation, Queen,
CloudAssignment, ...
The other domain classes are considered normal planning facts,
for example Shift, Employee, Bed, Room, Department, ...
They do not change during planning (at least not without pausing
the solver).
Read the manual to understand the "planning entity" concept better.
[MAJOR] You need to define your planning entity class(es) in the
configuration now:
Before in *SolverConfig.xml and *BenchmarkConfig.xml:
<localSearchSolver>
<solutionClass>....</solutionClass>
<scoreDrl>...</scoreDrl>
After in *SolverConfig.xml and *BenchmarkConfig.xml:
<localSearchSolver>
<solutionClass>....</solutionClass>
<planningEntityClass>org.drools.planner.examples.curriculumcourse.domain.Lecture</planningEntityClass>
<scoreDrl>...</scoreDrl>
[MAJOR] You need to annotate your planning entity class(es) with
the @PlanningEntity annotation
Before in *.java:
public class Lecture ... {
...
}
After in *.java:
@PlanningEntity
public class Lecture ... {
...
}
[RECOMMENDED] Understand the concept of a "planning variable"
property.
The property (or properties) on a planning entity class that are
changed (through their setter) during planning
are planning variables.
For example: ShiftAssignment.getEmployee(),
BedDesignation.getBed(), Queen.getY(), ...
Note that most planning entities have 1 property which defines the
planning entity
and that property is NOT a planning variable.
For example: ShiftAssignment.getShift(),
BedDesignation.getAdmissionPart(), Queen.getX(), ...
Read the manual to understand the "planning variable" concept better.
[MAJOR] You need to annotate your planning variable property(ies)
with the @PlanningVariable annotation.
Furthermore, you need to annotate a @ValueRange* annotation on to
define the allowed values.
Commonly, you 'll use @ValueRangeFromSolutionProperty which
specifies a property name on the solution
which returns a collection of the allowed values for that variable.
Before in *.java:
@PlanningEntity
public class Lecture ... {
private Course course;
private int lectureIndexInCourse;
// Changed by moves, between score calculations.
private Period period;
private Room room;
public Course getCourse() {...}
public void setCourse(Course course) {...}
public int getLectureIndexInCourse() {...}
public void setLectureIndexInCourse(int
lectureIndexInCourse) {...}
public Period getPeriod() {...}
public void setPeriod(Period period) {...}
public Room getRoom() {...}
public void setRoom(Room room) {...}
...
public int getStudentSize() {
return course.getStudentSize();
}
public Day getDay() {
return period.getDay();
}
}
After in *.java:
@PlanningEntity
public class Lecture ... {
private Course course;
private int lectureIndexInCourse;
// Changed by moves, between score calculations.
private Period period;
private Room room;
// This is not a PlanningVariable: it defines the planning
entity
public Course getCourse() {...}
public void setCourse(Course course) {...}
// This is not a PlanningVariable: it defines the planning
entity
public int getLectureIndexInCourse() {...}
public void setLectureIndexInCourse(int
lectureIndexInCourse) {...}
@PlanningVariable
@ValueRangeFromSolutionProperty(propertyName = "periodList")
public Period getPeriod() {...}
public void setPeriod(Period period) {...}
@PlanningVariable
@ValueRangeFromSolutionProperty(propertyName = "roomList")
public Room getRoom() {...}
public void setRoom(Room room) {...}
...
// This is not a PlanningVariable: no setter
public int getStudentSize() {
return course.getStudentSize();
}
// This is not a PlanningVariable: no setter
public Day getDay() {
return period.getDay();
}
}
[MAJOR] Annotate every property on your Solution that returns a
collection of planning entities
with @PlanningEntityCollectionProperty.
Before in *.java:
public class CurriculumCourseSchedule ... implements
Solution<...> {
private List<Lecture> lectureList;
...
public List<Lecture> getLectureList() {...}
public void setLectureList(List<Lecture> lectureList) {...}
}
After in *.java:
public class CurriculumCourseSchedule ... implements
Solution<...> {
private List<Lecture> lectureList;
...
@PlanningEntityCollectionProperty
public List<Lecture> getLectureList() {...}
public void setLectureList(List<Lecture> lectureList) {...}
}
[MAJOR] The method getFacts() has been removed from the Solution
interface.
Annotate every property that returns a fact or fact collection
with the @PlanningFactProperty
or @PlanningFactCollectionProperty annotation respectively,
except those already annotated with @PlanningEntityCollectionProperty.
Properties annotated with these annotations are inserted into the
working memory as facts:
- @PlanningFactProperty
- @PlanningFactCollectionProperty: each element in the collection
- @PlanningEntityCollectionProperty: each planning entity in the
collection that is initialized
Remove the getFacts() method.
Before in *.java:
public class ... implements Solution<...> {
private InstitutionalWeighting institutionalWeighting;
private List<Teacher> teacherList;
private List<Curriculum> curriculumList;
...
private List<UnavailablePeriodConstraint>
unavailablePeriodConstraintList;
private List<Lecture> lectureList;
private HardAndSoftScore score;
...
public String getName() {...}
public InstitutionalWeighting getInstitutionalWeighting()
{...}
public List<Teacher> getTeacherList() {...}
public List<Curriculum> getCurriculumList() {...}
...
public List<UnavailablePeriodConstraint>
getUnavailablePeriodConstraintList() {...}
@PlanningEntityCollectionProperty
public List<Lecture> getLectureList() {...}
public ...Score getScore() {...}
public Collection<? extends Object> getFacts() {
List<Object> facts = new ArrayList<Object>();
facts.addAll(teacherList);
facts.addAll(curriculumList);
...
facts.addAll(unavailablePeriodConstraintList);
if (isInitialized()) {
facts.addAll(lectureList);
}
facts.addAll(calculateTopicConflictList());
return facts;
}
public List<TopicConflict> calculateTopicConflictList() {...}
}
After in *.java:
public class ... implements Solution<...> {
private InstitutionalWeighting institutionalWeighting;
private List<Teacher> teacherList;
private List<Curriculum> curriculumList;
...
private List<UnavailablePeriodConstraint>
unavailablePeriodConstraintList;
private List<Lecture> lectureList;
private HardAndSoftScore score;
...
// This is not a PlanningFactProperty: the name is
inserted into the working memory
public String getName() {...}
@PlanningFactProperty
public InstitutionalWeighting getInstitutionalWeighting()
{...}
@PlanningFactCollectionProperty
public List<Teacher> getTeacherList() {...}
@PlanningFactCollectionProperty
public List<Curriculum> getCurriculumList() {...}
...
@PlanningFactCollectionProperty
public List<UnavailablePeriodConstraint>
getUnavailablePeriodConstraintList() {...}
// This is not a PlanningFactCollectionProperty: it is a
PlanningEntityCollectionProperty
@PlanningEntityCollectionProperty
public List<Lecture> getLectureList() {...}
// This is not a PlanningFactProperty: the score is
inserted into the working memory
public ...Score getScore() {...}
// renamed from calculateTopicConflictList because these
are also facts needed in the working memory
@PlanningFactCollectionProperty
public List<TopicConflict> getTopicConflictList() {...}
}
[RECOMMEND] A planning entity is considered uninitialized if one
if at least on of its planning variables is null.
Therefor it's now possible to start from a partially initialized
solution,
for example during real-time re-planning as new facts events come in.
[MAJOR] The StartingSolutionInitializer no longer has a
isSolutionInitialized(AbstractSolverScope) method
Before in *StartingSolutionInitializer.java:
public class ...StartingSolutionInitializer extends
AbstractStartingSolutionInitializer {
@Override
public boolean isSolutionInitialized(AbstractSolverScope
abstractSolverScope) {
...
}
...
}
After in *StartingSolutionInitializer.java:
public class ...StartingSolutionInitializer extends
AbstractStartingSolutionInitializer {
...
}
[MAJOR] The planning entity collection in the Solution can never
be null,
but some (or all) of its planning entity's can be uninitialized.
So create them before setting the starting solution, instead of in
your StartingSolutionInitializer.
Before in *.java:
public class ... {
public void ...() {
CurriculumCourseSchedule schedule = new
CurriculumCourseSchedule();
schedule.setTeacherList(teacherList);
schedule.setCourseList(courseList);
...
solver.setStartingSolution(schedule);
}
}
After in *.java:
public class ... {
public void ...() {
CurriculumCourseSchedule schedule = new
CurriculumCourseSchedule();
schedule.setTeacherList(teacherList);
schedule.setCourseList(courseList);
...
schedule.setLectureList(createLectureList(schedule.getCourseList()));
solver.setStartingSolution(schedule);
}
private List<Lecture> createLectureList(List<Course>
courseList) {
List<Lecture> lectureList = new
ArrayList<Lecture>(courseList.size());
long id = 0L;
for (Course course : courseList) {
for (int i = 0; i < course.getLectureSize(); i++) {
Lecture lecture = new Lecture();
lecture.setId((long) id);
id++;
lecture.setCourse(course);
// Make sure to set all non PlanningVariable
properties
lecture.setLectureIndexInCourse(i);
// Notice that we lave the PlanningVariable
properties on null
lectureList.add(lecture);
}
}
return lectureList;
}
}
[RECOMMENDED] Remove the isInitialized() from Solution if you
copied that from the examples.
Before in *.java:
public class ... implements Solution<...> {
public boolean isInitialized() {
return (lectureList != null);
}
...
}
After in *.java:
public class ... implements Solution<...> {
...
}
*
The Good News*
I already have written a brute force solver (only useful for very
very small toy problems of course).
In time, I 'll write a branch and bound solver (only useful for
very small toy problems of course).
I 'll write generic, high-quality StartingSolutionInitializers
that work on any program,
such as First Fit Decreasing and Cheapest Insertion.
A good StartingSolutionInitializer is really important to get a
good result from Planner.
Currently writing a StartingSolutionInitializer was a bit of a
black art.
Some users use a highly under optimized version or - even worse -
none at all.
Phasing. This will be fun.
You 'll be able to do configure this really easily:
Phase 1 = First Fit Decreasing (for initialization)
Phase 2 = Simulated Annealing (after initialization)
Phase 3 = Tabu search (when things get really hard)
Notice that a StartingSolutionInitializer is just a phase.
[1] https://github.com/droolsjbpm/drools-planner/pull/2/files
--
With kind regards,
Geoffrey De Smet
_______________________________________________
rules-dev mailing list
rules-dev@lists.jboss.org <mailto:rules-dev@lists.jboss.org>
https://lists.jboss.org/mailman/listinfo/rules-dev
_______________________________________________
rules-dev mailing list
rules-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-dev
--
With kind regards,
Geoffrey De Smet
_______________________________________________
rules-dev mailing list
rules-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-dev
