Changes in StandardGeneticAlgorithm: Freeze targets per generation; add GASolutionSource #1347

francastagna · 2025-10-15T16:22:22Z

PR Summary

We implemented changes to the StandardGeneticAlgorithm:

With and without archive: allows selecting the source of the final solution.
Fitness update: introduces generation-frozen objectives (“frozen targets”).

We implemented changes to the AbstractGeneticAlgorithm:

GA Recorder/Observer: adds observer hooks to record GA events (selection, crossover and mutation events).

Main Changes

Configuration: added EMConfig.GASolutionSource and EMConfig.gaSolutionSource to choose the source of the final solution: ARCHIVE or POPULATION.
Base GA: rewrote the core logic in AbstractGeneticAlgorithm (population, elitism, tournament, mutate, xover, sampleSuite).
score function: computes fitness using generation-frozen objectives (frozenTargets).
Dual support: different behavior when the final solution is built from the archive vs. from the population.
StandardGeneticAlgorithm.searchOnce():
- Freezes objectives at the start of each generation: frozenTargets = archive.notCoveredTargets().
- Builds the next population via tournament selection, crossover, and mutation, respecting the configured probabilities.
GA Recorder/Observer:
- Introduces observer hooks in the GA core to record operator metrics.
- Enhances debugging and testing capabilities without altering the GA behaviour.

Fitness Update and Use of `frozenTargets`

Why

We align with Campos’ paper recommendation to keep objectives fixed within a generation so that all individuals compete under the same set of pending goals.
This prevents fitness drift when the objective set changes mid-generation (due to new coverages), which would make comparisons between individuals inconsistent.

How

At the start of each generation, capture frozenTargets = archive.notCoveredTargets().
The score of a test suite is computed only over those frozen objectives throughout the generation.
At the end of the generation (or at a snapshot), the set can be updated.

Advantages

Intra-generation consistency for tournament/selection.
Fair and stable comparisons; avoids individual evaluation depending on the exact timing of its assessment.

…buildSolution; support archive vs population

… javadoc added

Changes in StandardGeneticAlgorithm: Freeze targets per generation; add GASolutionSource.

arcuri82

@fncastagna thx for this first PR ;) see my comments

arcuri82 · 2025-10-16T08:46:43Z

core/src/main/kotlin/org/evomaster/core/search/algorithms/strategy/DefaultMutationOperator.kt

+ * - add: if size < maxSearchSuiteSize, sample + evaluate; add to suite (and archive GASolutionSource = ARCHIVE).
+ * - mod: mutate one random element; re-evaluate (or mutateAndSave if GASolutionSource = ARCHIVE).
+ */
+class DefaultMutationOperator : MutationOperator {


the name might be misleading, as mutation here is just at test suite level and not test case level, isn't it? if so, either "suite" should be in the package name, or class name, somehow

You're right it can be misleading, I'll add suite to the package name.

arcuri82 · 2025-10-16T08:48:17Z

core/src/main/kotlin/org/evomaster/core/search/algorithms/strategy/DefaultMutationOperator.kt

+        private const val OP_MOD = "mod"
+    }
+
+    override fun <T : Individual> mutateIndividual(


i don't like the name of this function. this is doing a lot of things, and not just "mutate". name should be more descriptive

however, looking at the old code, it was named like that as well... :) but i guess it is a good time now for a name update.

arcuri82 · 2025-10-16T08:49:27Z

core/src/main/kotlin/org/evomaster/core/search/algorithms/strategy/MutationOperator.kt

+import org.evomaster.core.search.service.Sampler
+import org.evomaster.core.search.service.mutator.Mutator
+
+interface MutationOperator {


is this really a mtation operator? you are also evaluating fitness, and saving to archive based on that

arcuri82 · 2025-10-16T08:51:46Z

...src/main/kotlin/org/evomaster/core/search/algorithms/strategy/TournamentSelectionStrategy.kt

+        score: (WtsEvalIndividual<T>) -> Double
+    ): WtsEvalIndividual<T> {
+        val selected = randomness.choose(population, tournamentSize)
+        return selected.maxByOrNull { score(it) } ?: randomness.choose(population)


how can this be null? would that be a bug, or something that could actually happen? here, if it happens it silently handled

core/src/main/kotlin/org/evomaster/core/search/algorithms/AbstractGeneticAlgorithm.kt

arcuri82 · 2025-10-16T09:20:31Z

core/src/main/kotlin/org/evomaster/core/EMConfig.kt

+    /**
+     * Source to build the final GA solution.
+     * ARCHIVE: use current behavior (take tests from the archive).
+     * POPULATION: for GA algorithms, take the best individual from the final population.


clarify here when evolving full test suites, and not tests in isolation

arcuri82 · 2025-10-16T09:23:40Z

core/src/test/kotlin/org/evomaster/core/search/algorithms/StandardGeneticAlgorithmTest.kt


 class StandardGeneticAlgorithmTest {

    val injector: Injector = LifecycleInjector.builder()


if you are having more than 1 test in this class, this is problematic. shared injector creates dependencies. should be recreated for each test, eg, in a @BeforeEach method

arcuri82 · 2025-10-16T09:24:13Z

core/src/test/kotlin/org/evomaster/core/search/algorithms/StandardGeneticAlgorithmTest.kt


            val epc = injector.getInstance(ExecutionPhaseController::class.java)
+            if (epc.isInSearch()) {
+                epc.finishSearch()


not needed if injector is recreated per test

arcuri82 · 2025-10-16T09:24:55Z

core/src/test/kotlin/org/evomaster/core/search/algorithms/StandardGeneticAlgorithmTest.kt

-            val solution = standardGeneticAlgorithm.search()
-
+                standardGeneticAlgorithm.search()
+            } finally {


likely not needed if injector is recreated per test

n EMConfig

…e, translate spanish comments to english

…rithm-2 Feature/standard genetic algorithm

fncastagna and others added 13 commits October 6, 2025 09:37

feat(core/ga): freeze targets per generation; add GASolutionSource + …

d313f00

…buildSolution; support archive vs population

fix cp error in StandardGeneticAlgorithmTest.kt

8fb5bbf

add tests to StandardGeneticAlgorithmTest.kt

8288e69

added publisher observer pattern

b5e7283

rename crossover and mutator operator method names

916290a

applyMutation renamed to mutateIndividual and DefaultCrossoveOperator…

23c0060

… javadoc added

added DefaultMutationOperator javadoc

f8cd208

added TournamentSelectionStrategy javadocs

d111925

fix matching brackets

19f5c41

add tournament selection strategy javadocs

08e98b0

deleted accidentaly commited file MonotonicGeneticAlgorithmTest

a0dc76c

Merge pull request #3 from francastagna/feature/genetics-algorithms

0259fe2

Changes in StandardGeneticAlgorithm: Freeze targets per generation; add GASolutionSource.

Merge branch 'master' into master

46e4164

jgaleotti requested a review from arcuri82 October 15, 2025 16:30

arcuri82 requested changes Oct 16, 2025

View reviewed changes

fncastagna added 6 commits October 17, 2025 08:13

remove GA bindings from BaseModule

8749385

improve explanation of GASolutionSource i

7bcac3b

n EMConfig

remove injection of GA operators, changed inline for loop to multilin…

7edce0d

…e, translate spanish comments to english

added suite to GA operators package name

e931072

added @beforeeach to tests in StandardGeneticAlgorithmTest

14ffc21

added suite to GA operators package name

28d9e72

francastagna mentioned this pull request Oct 17, 2025

Feature/standard genetic algorithm francastagna/EvoMaster#7

Merged

fncastagna and others added 4 commits October 17, 2025 08:48

minor fixes

40869f9

compilation and test fixes

4c82ab9

renamed mutation operator

013a774

Merge pull request #7 from francastagna/feature/standard-genetic-algo…

7bf221f

…rithm-2 Feature/standard genetic algorithm

francastagna requested a review from arcuri82 October 17, 2025 16:53

arcuri82 approved these changes Oct 20, 2025

View reviewed changes

arcuri82 changed the base branch from master to pr-remote-francastagna October 20, 2025 11:12

arcuri82 merged commit 3efeb01 into WebFuzzing:pr-remote-francastagna Oct 20, 2025

arcuri82 mentioned this pull request Oct 20, 2025

Pr remote francastagna #1351

Merged


		class StandardGeneticAlgorithmTest {

		val injector: Injector = LifecycleInjector.builder()

Changes in StandardGeneticAlgorithm: Freeze targets per generation; add GASolutionSource #1347

Changes in StandardGeneticAlgorithm: Freeze targets per generation; add GASolutionSource #1347

Uh oh!

Conversation

francastagna commented Oct 15, 2025

PR Summary

Main Changes

Fitness Update and Use of frozenTargets

Why

How

Advantages

Uh oh!

arcuri82 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Fitness Update and Use of `frozenTargets`