Cellular GA

[francastagna]

PR Summary

We implemented and validated the Cellular Genetic Algorithm (cGA) and added comprehensive tests.

---

Cellular GA Pseudocode

Source: https://www.researchgate.net/publication/327166354_An_Empirical_Evaluation_of_Evolutionary_Algorithms_for_Unit_Test_Suite_Generation

---

Main Changes

• New Algorithm (CellularGeneticAlgorithm):

  • Performs neighborhood-restricted evolution: for each index i, selects two parents from the neighborhood, produces 2 offspring, and keeps the local best between parent and offspring.
  • Invokes lifecycle hooks: beginGeneration, endGeneration, beginStep, endStep.
  • Freezes targets each iteration: frozenTargets = archive.notCoveredTargets().

• Config (EMConfig):

  • Added CellularGA to EMConfig.Algorithm.
  • Added CGANeighborhoodModel { RING, L5, C9, C13 } and property cgaNeighborhoodModel (default RING).

---

Tests

• Neighborhood tests added (NeighborhoodTest.kt):

  • Validates index wrap-around and neighborhood composition for RING/L5/C9/C13 using identity checks.

• Cellular GA tests added (CellularGeneticAlgorithmTest.kt):

  • Finds optimum on OneMax (testCellularGeneticAlgorithmFindsOptimum).
  • Next generation shape (testNextGenerationIsLocalWinnersFromNeighborhood):
    • Preserves population size; for each position i, next[i] is either the original parent or one of the two offspring produced at i.
  • Edge cases:
    • xoverProbability = 0.0, fixedRateMutation = 1.0 → no crossovers; one mutation per position; size preserved.
    • fixedRateMutation = 0.0, xoverProbability = 1.0 → one crossover per position; zero mutations; size preserved.

---