n-Way Partitioning
Hard problem and no really good heuristics for nɮ
- Direct Methods: Start with seed node for each partition and assign nodes to each partition using some criterion (e.g. sum of weighted connections into partition)
- Group Migration Methods: Start with (random) initial partition and migrate nodes among partitions via some heuristic
- Metric Allocation Methods: uses metrics other than connection graph and clusters nodes based on metric.
- Stochastic Optimization Approaches: Use a general-purpose stochastic approach like simulated annealing or genetic algorithms
Usually apply two-way partitioning (Kernighan-Lin or Fiduccia-Matheyses) recursively, or simulated annealing