Release Notes 6.2

We are happy to announce a 6.2 Final release of OptaPlanner. OptaPlanner is a lightweight, embeddable planning engine written in Java™ to solve AI constraint optimization problems efficiently. Use cases include Vehicle Routing, Employee Rostering, Maintenance Scheduling, Task Assignment, School Timetabling, Cloud Optimization, Conference Scheduling and many more.

Download
9.38.0.Final

Documentation
9.38.0.Final

Status of OptaPlanner

Stable: Lots of unit, integration and stress tests
Reliable: Used across the world in production
Scalable: To billions of constraint matches with minimal RAM and CPU time
Documented: Read the detailed reference manual and the many examples
Open Source: Apache License 2.0

Note for Red Hat Decision Manager customers

The RHDM version differs from the OptaPlanner version:

RHDM version	OptaPlanner version
7.8	7.39
7.9	7.44
7.1	7.48
7.11	8.5 (and 7.52)
7.12	8.11 (and 7.59)
7.13	8.13 (and 7.67)

New and noteworthy

Scalable VRP with nearby selection

Nearby selection allows a vehicle routing problem to scale out gracefully beyond 1000 locations, without the need for partitioning. It works by focusing on move selections that modify locations that are near each other:

Nearby selection random distribution

It results in much better scalability on larger datasets, for example a VRP with 2750 customers (higher is better), which reduces fuel and labor expenses by 15% in a 5 minute run:

Nearby selection random distribution

Several nearby selection probability distributions are supported: block distribution, linear distribution, parabolic distribution and beta distribution.

TailChainSwapMove (2-opt) for VRP

TailChainSwapMove is a new move type for chained variables. It’s a subset of SubchainChangeMove and SubchainSwapMove, but it’s generally more efficient, especially for time windowed cases.

In our benchmarks, a union of ChangeMove, SwapMove and TailChainSwapMove (using nearby selection on all 3) performed best.

Improved built-in variable listener efficiency

VRP with a @InverseRelationShadowVariable is now more efficient. In some cases, it’s up to 32% faster.

Strategic Oscillation Tabu Search

Strategic Oscillation Tabu Search is often an improvement over normal Tabu Search. Instead of picking the accepted move with the highest score, it employs a different mechanism: If there’s an improving move, it picks it. If there’s no improving move however, it prefers moves which improve a softer score level, over moves which break a harder score level less.

To enable it, do this:

  <localSearch>
    ...
    <acceptor>
      <entityTabuSize>7</entityTabuSize>
    </acceptor>
    <forager>
      <acceptedCountLimit>1000</acceptedCountLimit>
      <finalistPodiumType>STRATEGIC_OSCILLATION</finalistPodiumType>
    </forager>
  </localSearch>

New example: Cheap time scheduling

Schedule all tasks in time and on a machine to minimize the power cost. Each machine must have enough hardware to run all of its tasks. Each task and machine consumes power. The power price differs over time.

Cheap time example

Based on contributions by Lukáš Petrovický.

New benchmarker statistics: Constraint Match Total Best/Step score

These new statistics visualize how the individual constraint types change over time.

constraint match total best score statistic

This gives a better insight as to which constraints impact the score the most.

Other improvements

Construction Heuristics: new pick early type: FIRST_FEASIBLE_SCORE which is useful for scaling.
Benchmarker: logarithmic scale for Problem scale axis when appropriate. Contributed by Ondrej Skopek.
BendableLongScore: Bendable score with long types. Contributed by Dieter De Paepe.