c1a44fd2a6
Add README.md files for moeo/tutorial/Lesson{1-4}, smp/tutorial/Lesson{1-4},
and mo/tutorial/Lesson9 — these tutorials had no documentation.
Fix return 1 → return 0 in 28 tutorial main() functions across mo/ and
smp/ that unconditionally returned failure status.
48 lines
1.4 KiB
Markdown
48 lines
1.4 KiB
Markdown
# Variable Neighborhood Search (VNS)
|
|
|
|
VNS on the N-Queens problem, using multiple neighborhood structures to
|
|
escape local optima.
|
|
|
|
The N-Queens problem places N queens on an N x N chessboard so that no two
|
|
queens attack each other. The representation is a permutation (one queen
|
|
per column) and the fitness counts conflicts (minimized).
|
|
|
|
## Running
|
|
|
|
From the `build/mo/tutorial/Lesson9` directory:
|
|
|
|
```shell
|
|
./VNS -V=12
|
|
```
|
|
|
|
This runs VNS on a 12-queens instance with a 3-second time limit.
|
|
|
|
## How it works
|
|
|
|
VNS alternates between perturbation (shaking) and local search using
|
|
different neighborhood structures. Here, two neighborhoods are used:
|
|
|
|
```c++
|
|
shiftNeighborhood shiftNH((vecSize-1) * (vecSize-1));
|
|
swapNeighborhood swapNH(vecSize * (vecSize-1) / 2);
|
|
```
|
|
|
|
Each has its own hill-climber and mutation operator:
|
|
|
|
```c++
|
|
moSimpleHC<shiftNeighbor> ls1(shiftNH, fullEval, shiftEval);
|
|
moSimpleHC<swapNeighbor> ls2(swapNH, fullEval, swapEval);
|
|
|
|
moRndVectorVNSelection<Queen> selectNH(ls1, shiftMut, true);
|
|
selectNH.add(ls2, swapMut);
|
|
```
|
|
|
|
`moRndVectorVNSelection` randomly picks which neighborhood/local search
|
|
to apply at each iteration. Forward (`moForwardVectorVNSelection`) and
|
|
backward (`moBackwardVectorVNSelection`) selection strategies are also
|
|
available. The search runs until the `moTimeContinuator` time limit:
|
|
|
|
```c++
|
|
moTimeContinuator<shiftNeighbor> cont(3); // 3 seconds
|
|
moVNS<shiftNeighbor> vns(selectNH, acceptCrit, fullEval, cont);
|
|
```
|