HardMediumSoftMicrosScore

[ge0ffrey]

Motivation:

In use cases with a fairness or load balancing constraint, alongside other soft constraints, the only practical way to write that constraint is using the sqrt(sum(x²)) pattern. In the tennis example there's a loadBalance() function that does that for you.

This sqrt() gives a decimal number and that does not work well with LongScore:

• BigDecimalScore is slow and a bit of a pain
• Only the first 4 decimals matter (simple math support that this accuracy is accurate enough). So typically it is mulitplied by 1 milliion and put into a LongScore. The 6 digits (micros) are 2 digits for safety.

However, this is a pain:

• for clarity (think logs etc)
• for visualization in any UI (including score analysis etc)
• for score weights

Proposal A

A new score: HardMediumSoftMicrosScore. It uses long underneath.

• long hardMicros
• long mediumMicros
• long softMicros
• toString() = hardMicros / 1M + (hardMicros % 1M == 0 ? "" : hardMicros % 1M ...
• ONE_SOFT => softMicros = 1_000_000
• ofSoft(7) => softMicros = 7_000_000
• ofSoftMicros(7) => softMicros = 7

The score weight of the dayOffRequest constraint is ofSoft(10).

Is the score weight of the fairness constraint is ofSoftMicros(1)?

• A1) Yes. But that's confusing for users to tune.
• A2) No, it's ofSoft(1) and the constraint impl does a trick so the accuracy isn't lost.

Related design

• [ ] Standarize and document loadBalance() of the tennis example
• [ ] Guide for fairness and load balancing, including FTE (Full time equivelant) challenges
• [ ] An understandable version of math of https://www.optaplanner.org/blog/2017/02/03/FormulaForMeasuringUnfairness.html