Dieser Inhalt ist in der von Ihnen ausgewählten Sprache nicht verfügbar.
Chapter 14. Red Hat Build of OptaPlanner and Java: a school timetable quickstart guide
This guide walks you through the process of creating a simple Java application with the OptaPlanner constraint solving artificial intelligence (AI). You will build a command-line application that optimizes a school timetable for students and teachers:
Your application will assign Lesson instances to Timeslot and Room instances automatically by using AI to adhere to hard and soft scheduling constraints, for example:
- A room can have at most one lesson at the same time.
- A teacher can teach at most one lesson at the same time.
- A student can attend at most one lesson at the same time.
- A teacher prefers to teach all lessons in the same room.
- A teacher prefers to teach sequential lessons and dislikes gaps between lessons.
- A student dislikes sequential lessons on the same subject.
Mathematically speaking, school timetabling is an NP-hard problem. This means it is difficult to scale. Simply brute force iterating through all possible combinations takes millions of years for a non-trivial data set, even on a supercomputer. Fortunately, AI constraint solvers such as OptaPlanner have advanced algorithms that deliver a near-optimal solution in a reasonable amount of time.
Prerequisites
- OpenJDK (JDK) 11 is installed. Red Hat build of Open JDK is available from the Software Downloads page in the Red Hat Customer Portal (login required).
- Apache Maven 3.6 or higher is installed. Maven is available from the Apache Maven Project website.
- An IDE, such as IntelliJ IDEA, VSCode or Eclipse
14.1. Creating the Maven or Gradle build file and add dependencies Link kopierenLink in die Zwischenablage kopiert!
You can use Maven or Gradle for the OptaPlanner school timetable application. After you create the build files, add the following dependencies:
-
optaplanner-core(compile scope) to solve the school timetable problem -
optaplanner-test(test scope) to JUnit test the school timetabling constraints -
An implementation such as
logback-classic(runtime scope) to view the steps that OptaPlanner takes
Procedure
- Create the Maven or Gradle build file.
Add
optaplanner-core,optaplanner-test, andlogback-classicdependencies to your build file:For Maven, add the following dependencies to the
pom.xmlfile:Copy to Clipboard Copied! Toggle word wrap Toggle overflow The following example shows the complete
pom.xmlfile.Copy to Clipboard Copied! Toggle word wrap Toggle overflow For Gradle, add the following dependencies to the
gradle.buildfile:Copy to Clipboard Copied! Toggle word wrap Toggle overflow The following example shows the completed
gradle.buildfile.Copy to Clipboard Copied! Toggle word wrap Toggle overflow
14.2. Model the domain objects Link kopierenLink in die Zwischenablage kopiert!
The goal of the Red Hat Build of OptaPlanner timetable project is to assign each lesson to a time slot and a room. To do this, add three classes, Timeslot, Lesson, and Room, as shown in the following diagram:
Timeslot
The Timeslot class represents a time interval when lessons are taught, for example, Monday 10:30 - 11:30 or Tuesday 13:30 - 14:30. In this example, all time slots have the same duration and there are no time slots during lunch or other breaks.
A time slot has no date because a high school schedule just repeats every week. There is no need for continuous planning. A timeslot is called a problem fact because no Timeslot instances change during solving. Such classes do not require any OptaPlanner-specific annotations.
Room
The Room class represents a location where lessons are taught, for example, Room A or Room B. In this example, all rooms are without capacity limits and they can accommodate all lessons.
Room instances do not change during solving so Room is also a problem fact.
Lesson
During a lesson, represented by the Lesson class, a teacher teaches a subject to a group of students, for example, Math by A.Turing for 9th grade or Chemistry by M.Curie for 10th grade. If a subject is taught multiple times each week by the same teacher to the same student group, there are multiple Lesson instances that are only distinguishable by id. For example, the 9th grade has six math lessons a week.
During solving, OptaPlanner changes the timeslot and room fields of the Lesson class to assign each lesson to a time slot and a room. Because OptaPlanner changes these fields, Lesson is a planning entity:
Most of the fields in the previous diagram contain input data, except for the orange fields. A lesson’s timeslot and room fields are unassigned (null) in the input data and assigned (not null) in the output data. OptaPlanner changes these fields during solving. Such fields are called planning variables. In order for OptaPlanner to recognize them, both the timeslot and room fields require an @PlanningVariable annotation. Their containing class, Lesson, requires an @PlanningEntity annotation.
Procedure
Create the
src/main/java/com/example/domain/Timeslot.javaclass:Copy to Clipboard Copied! Toggle word wrap Toggle overflow Notice the
toString()method keeps the output short so it is easier to read OptaPlanner’sDEBUGorTRACElog, as shown later.Create the
src/main/java/com/example/domain/Room.javaclass:Copy to Clipboard Copied! Toggle word wrap Toggle overflow Create the
src/main/java/com/example/domain/Lesson.javaclass:Copy to Clipboard Copied! Toggle word wrap Toggle overflow The
Lessonclass has an@PlanningEntityannotation, so OptaPlanner knows that this class changes during solving because it contains one or more planning variables.The
timeslotfield has an@PlanningVariableannotation, so OptaPlanner knows that it can change its value. In order to find potentialTimeslotinstances to assign to this field, OptaPlanner uses thevalueRangeProviderRefsproperty to connect to a value range provider that provides aList<Timeslot>to pick from. See Section 14.4, “Gather the domain objects in a planning solution” for information about value range providers.The
roomfield also has an@PlanningVariableannotation for the same reasons.
14.3. Define the constraints and calculate the score Link kopierenLink in die Zwischenablage kopiert!
When solving a problem, a score represents the quality of a specific solution. The higher the score the better. Red Hat Build of OptaPlanner looks for the best solution, which is the solution with the highest score found in the available time. It might be the optimal solution.
Because the timetable example use case has hard and soft constraints, use the HardSoftScore class to represent the score:
- Hard constraints must not be broken. For example: A room can have at most one lesson at the same time.
- Soft constraints should not be broken. For example: A teacher prefers to teach in a single room.
Hard constraints are weighted against other hard constraints. Soft constraints are weighted against other soft constraints. Hard constraints always outweigh soft constraints, regardless of their respective weights.
To calculate the score, you could implement an EasyScoreCalculator class:
Unfortunately, this solution does not scale well because it is non-incremental: every time a lesson is assigned to a different time slot or room, all lessons are re-evaluated to calculate the new score.
A better solution is to create a src/main/java/com/example/solver/TimeTableConstraintProvider.java class to perform incremental score calculation. This class uses OptaPlanner’s ConstraintStream API which is inspired by Java 8 Streams and SQL. The ConstraintProvider scales an order of magnitude better than the EasyScoreCalculator: O(n) instead of O(n²).
Procedure
Create the following src/main/java/com/example/solver/TimeTableConstraintProvider.java class:
14.4. Gather the domain objects in a planning solution Link kopierenLink in die Zwischenablage kopiert!
A TimeTable instance wraps all Timeslot, Room, and Lesson instances of a single dataset. Furthermore, because it contains all lessons, each with a specific planning variable state, it is a planning solution and it has a score:
-
If lessons are still unassigned, then it is an uninitialized solution, for example, a solution with the score
-4init/0hard/0soft. -
If it breaks hard constraints, then it is an infeasible solution, for example, a solution with the score
-2hard/-3soft. -
If it adheres to all hard constraints, then it is a feasible solution, for example, a solution with the score
0hard/-7soft.
The TimeTable class has an @PlanningSolution annotation, so Red Hat Build of OptaPlanner knows that this class contains all of the input and output data.
Specifically, this class is the input of the problem:
A
timeslotListfield with all time slots- This is a list of problem facts, because they do not change during solving.
A
roomListfield with all rooms- This is a list of problem facts, because they do not change during solving.
A
lessonListfield with all lessons- This is a list of planning entities because they change during solving.
Of each
Lesson:-
The values of the
timeslotandroomfields are typically stillnull, so unassigned. They are planning variables. -
The other fields, such as
subject,teacherandstudentGroup, are filled in. These fields are problem properties.
-
The values of the
However, this class is also the output of the solution:
-
A
lessonListfield for which eachLessoninstance has non-nulltimeslotandroomfields after solving -
A
scorefield that represents the quality of the output solution, for example,0hard/-5soft
Procedure
Create the src/main/java/com/example/domain/TimeTable.java class:
The value range providers
The timeslotList field is a value range provider. It holds the Timeslot instances which OptaPlanner can pick from to assign to the timeslot field of Lesson instances. The timeslotList field has an @ValueRangeProvider annotation to connect those two, by matching the id with the valueRangeProviderRefs of the @PlanningVariable in the Lesson.
Following the same logic, the roomList field also has an @ValueRangeProvider annotation.
The problem fact and planning entity properties
Furthermore, OptaPlanner needs to know which Lesson instances it can change as well as how to retrieve the Timeslot and Room instances used for score calculation by your TimeTableConstraintProvider.
The timeslotList and roomList fields have an @ProblemFactCollectionProperty annotation, so your TimeTableConstraintProvider can select from those instances.
The lessonList has an @PlanningEntityCollectionProperty annotation, so OptaPlanner can change them during solving and your TimeTableConstraintProvider can select from those too.
14.5. The TimeTableApp.java class Link kopierenLink in die Zwischenablage kopiert!
After you have created all of the components of the school timetable application, you will put them all together in the TimeTableApp.java class.
The main() method performs the following tasks:
-
Creates the
SolverFactoryto build aSolverfor each data set. - Loads a data set.
-
Solves it with
Solver.solve(). - Visualizes the solution for that data set.
Typically, an application has a single SolverFactory to build a new Solver instance for each problem data set to solve. A SolverFactory is thread-safe, but a Solver is not. For the school timetable application, there is only one data set, so only one Solver instance.
Here is the completed TimeTableApp.java class:
The main() method first creates the SolverFactory:
The SolverFactory creation registers the @PlanningSolution class, the @PlanningEntity classes, and the ConstraintProvider class, all of which you created earlier.
Without a termination setting or a terminationEarly() event, the solver runs forever. To avoid that, the solver limits the solving time to five seconds.
After five seconds, the main() method loads the problem, solves it, and prints the solution:
The solve() method doesn’t return instantly. It runs for five seconds before returning the best solution.
OptaPlanner returns the best solution found in the available termination time. Due to the nature of NP-hard problems, the best solution might not be optimal, especially for larger data sets. Increase the termination time to potentially find a better solution.
The generateDemoData() method generates the school timetable problem to solve.
The printTimetable() method prettyprints the timetable to the console, so it’s easy to determine visually whether or not it’s a good schedule.
14.6. Creating and running the school timetable application Link kopierenLink in die Zwischenablage kopiert!
Now that you have completed all of the components of the school timetable Java application, you are ready to put them all together in the TimeTableApp.java class and run it.
Prerequisites
- You have created all of the required components of the school timetable application.
Procedure
Create the
src/main/java/org/acme/schooltimetabling/TimeTableApp.javaclass:Copy to Clipboard Copied! Toggle word wrap Toggle overflow Run the
TimeTableAppclass as the main class of a normal Java application. The following output should result:Copy to Clipboard Copied! Toggle word wrap Toggle overflow -
Verify the console output. Does it conform to all hard constraints? What happens if you comment out the
roomConflictconstraint inTimeTableConstraintProvider?
The info log shows what OptaPlanner did in those five seconds:
... Solving started: time spent (33), best score (-8init/0hard/0soft), environment mode (REPRODUCIBLE), random (JDK with seed 0). ... Construction Heuristic phase (0) ended: time spent (73), best score (0hard/0soft), score calculation speed (459/sec), step total (4). ... Local Search phase (1) ended: time spent (5000), best score (0hard/0soft), score calculation speed (28949/sec), step total (28398). ... Solving ended: time spent (5000), best score (0hard/0soft), score calculation speed (28524/sec), phase total (2), environment mode (REPRODUCIBLE).
... Solving started: time spent (33), best score (-8init/0hard/0soft), environment mode (REPRODUCIBLE), random (JDK with seed 0).
... Construction Heuristic phase (0) ended: time spent (73), best score (0hard/0soft), score calculation speed (459/sec), step total (4).
... Local Search phase (1) ended: time spent (5000), best score (0hard/0soft), score calculation speed (28949/sec), step total (28398).
... Solving ended: time spent (5000), best score (0hard/0soft), score calculation speed (28524/sec), phase total (2), environment mode (REPRODUCIBLE).
14.7. Testing the application Link kopierenLink in die Zwischenablage kopiert!
A good application includes test coverage. Test the constraints and the solver in your timetable project.
14.7.1. Test the school timetable constraints Link kopierenLink in die Zwischenablage kopiert!
To test each constraint of the timetable project in isolation, use a ConstraintVerifier in unit tests. This tests each constraint’s corner cases in isolation from the other tests, which lowers maintenance when adding a new constraint with proper test coverage.
This test verifies that the constraint TimeTableConstraintProvider::roomConflict, when given three lessons in the same room and two of the lessons have the same timeslot, penalizes with a match weight of 1. So if the constraint weight is 10hard it reduces the score by -10hard.
Procedure
Create the src/test/java/org/acme/optaplanner/solver/TimeTableConstraintProviderTest.java class:
Notice how ConstraintVerifier ignores the constraint weight during testing even if those constraint weights are hardcoded in the ConstraintProvider. This is because constraint weights change regularly before going into production. This way, constraint weight tweaking does not break the unit tests.
14.7.2. Test the school timetable solver Link kopierenLink in die Zwischenablage kopiert!
This example tests the Red Hat Build of OptaPlanner school timetable project on the Red Hat build of Quarkus platform. It uses a JUnit test to generate a test data set and send it to the TimeTableController to solve.
Procedure
Create the
src/test/java/com/example/rest/TimeTableResourceTest.javaclass with the following content:Copy to Clipboard Copied! Toggle word wrap Toggle overflow This test verifies that after solving, all lessons are assigned to a time slot and a room. It also verifies that it found a feasible solution (no hard constraints broken).
Add test properties to the
src/main/resources/application.propertiesfile:Copy to Clipboard Copied! Toggle word wrap Toggle overflow
Normally, the solver finds a feasible solution in less than 200 milliseconds. Notice how the application.properties file overwrites the solver termination during tests to terminate as soon as a feasible solution (0hard/*soft) is found. This avoids hard coding a solver time, because the unit test might run on arbitrary hardware. This approach ensures that the test runs long enough to find a feasible solution, even on slow systems. But it does not run a millisecond longer than it strictly must, even on fast systems.
14.8. Logging Link kopierenLink in die Zwischenablage kopiert!
After you complete the Red Hat Build of OptaPlanner school timetable project, you can use logging information to help you fine-tune the constraints in the ConstraintProvider. Review the score calculation speed in the info log file to assess the impact of changes to your constraints. Run the application in debug mode to show every step that your application takes or use trace logging to log every step and every move.
Procedure
- Run the school timetable application for a fixed amount of time, for example, five minutes.
Review the score calculation speed in the
logfile as shown in the following example:... Solving ended: ..., score calculation speed (29455/sec), ...
... Solving ended: ..., score calculation speed (29455/sec), ...Copy to Clipboard Copied! Toggle word wrap Toggle overflow -
Change a constraint, run the planning application again for the same amount of time, and review the score calculation speed recorded in the
logfile. Run the application in debug mode to log every step that the application makes:
-
To run debug mode from the command line, use the
-Dsystem property. To permanently enable debug mode, add the following line to the
application.propertiesfile:quarkus.log.category."org.optaplanner".level=debug
quarkus.log.category."org.optaplanner".level=debugCopy to Clipboard Copied! Toggle word wrap Toggle overflow The following example shows output in the
logfile in debug mode:... Solving started: time spent (67), best score (-20init/0hard/0soft), environment mode (REPRODUCIBLE), random (JDK with seed 0). ... CH step (0), time spent (128), score (-18init/0hard/0soft), selected move count (15), picked move ([Math(101) {null -> Room A}, Math(101) {null -> MONDAY 08:30}]). ... CH step (1), time spent (145), score (-16init/0hard/0soft), selected move count (15), picked move ([Physics(102) {null -> Room A}, Physics(102) {null -> MONDAY 09:30}]). ...... Solving started: time spent (67), best score (-20init/0hard/0soft), environment mode (REPRODUCIBLE), random (JDK with seed 0). ... CH step (0), time spent (128), score (-18init/0hard/0soft), selected move count (15), picked move ([Math(101) {null -> Room A}, Math(101) {null -> MONDAY 08:30}]). ... CH step (1), time spent (145), score (-16init/0hard/0soft), selected move count (15), picked move ([Physics(102) {null -> Room A}, Physics(102) {null -> MONDAY 09:30}]). ...Copy to Clipboard Copied! Toggle word wrap Toggle overflow
-
To run debug mode from the command line, use the
-
Use
tracelogging to show every step and every move for each step.
14.9. Using Micrometer and Prometheus to monitor your school timetable OptaPlanner Java application Link kopierenLink in die Zwischenablage kopiert!
OptaPlanner exposes metrics through Micrometer, a metrics instrumentation library for Java applications. You can use Micrometer with Prometheus to monitor the OptaPlanner solver in the school timetable application.
Prerequisites
- You have created the OptaPlanner school timetable application with Java.
- Prometheus is installed. For information about installing Prometheus, see the Prometheus website.
Procedure
Add the Micrometer Prometheus dependencies to the school timetable
pom.xmlfile where<MICROMETER_VERSION>is the version of Micrometer that you installed:<dependency> <groupId>io.micrometer</groupId> <artifactId>micrometer-registry-prometheus</artifactId> <version><MICROMETER_VERSION></version> </dependency>
<dependency> <groupId>io.micrometer</groupId> <artifactId>micrometer-registry-prometheus</artifactId> <version><MICROMETER_VERSION></version> </dependency>Copy to Clipboard Copied! Toggle word wrap Toggle overflow NoteThe
micrometer-coredependency is also required. However this dependency is contained in theoptaplanner-coredependency so you do not need to add it to thepom.xmlfile.Add the following import statements to the
TimeTableApp.javaclass.import io.micrometer.core.instrument.Metrics; import io.micrometer.prometheus.PrometheusConfig; import io.micrometer.prometheus.PrometheusMeterRegistry;
import io.micrometer.core.instrument.Metrics; import io.micrometer.prometheus.PrometheusConfig; import io.micrometer.prometheus.PrometheusMeterRegistry;Copy to Clipboard Copied! Toggle word wrap Toggle overflow Add the following lines to the top of the main method of the
TimeTableApp.javaclass so Prometheus can scrap data fromcom.sun.net.httpserver.HttpServerbefore the solution starts:Copy to Clipboard Copied! Toggle word wrap Toggle overflow Add the following line to control the solving time. By adjusting the solving time, you can see how the metrics change based on the time spent solving.
withTerminationSpentLimit(Duration.ofMinutes(5)));
withTerminationSpentLimit(Duration.ofMinutes(5)));Copy to Clipboard Copied! Toggle word wrap Toggle overflow - Start the school timetable application.
-
Open
http://localhost:8080/prometheusin a web browser to view the timetable application in Prometheus. Open your monitoring system to view the metrics for your OptaPlanner project.
The following metrics are exposed:
-
optaplanner_solver_errors_total: the total number of errors that occurred while solving since the start of the measuring. -
optaplanner_solver_solve_duration_seconds_active_count: the number of solvers currently solving. -
optaplanner_solver_solve_duration_seconds_max: run time of the longest-running currently active solver. -
optaplanner_solver_solve_duration_seconds_duration_sum: the sum of each active solver’s solve duration. For example, if there are two active solvers, one running for three minutes and the other for one minute, the total solve time is four minutes.
-