#152 Add minimum feedback set algorithms #165

jimbethancourt · 2025-12-18T17:05:13Z

Now calculating Minimum Feedback Arc Set instead of generating a DSM to identify relationships that need to be removed
Calculating Minimum Feedback Vertex Set to identify the direction relationships should move
Prioritizing relationships that need to be removed based on number of cycles they are in, the strength of the relationship, and the amount of churn of the source and target class
Fixed bug that reversed the relationship direction for generics

- Applied Spotless

Initial commit of SuperTypeToken. SuperTypeToken will make it possible to reify graphs in feedback algorithms when needed.

Now using SuperTypeToken in DirectedFeedbackVertexSetSolver and MinimumFeedbackArcSetSolver when creating new graphs in the algorithms

No longer using parallel streams to create vertices in createRandomGraph() to ensure that all vertices are created.

All MinimumFeedbackArcSetSolverTest unit tests are now passing or are disabled with good reason

Adding edges serially instead of in parallel in MinimumFeedbackArcSetSolver.createGraphWithoutEdges(). Adding edges in parallel was causing ArrayIndexOutOfBounds exceptions and appeared to run slightly slower as well when it did succeed. ClassCastException exceptions also stopped. Remediated several issues identified by sonar

No longer adding vertices in parallel. Adding vertices in parallel consistently led to ArrayOutOfBounds exceptions

…r method

Initial commit of eta (η) and k computation. Some unit tests are failing - will investigate further.

Initial commit of ModulatorComputer - includes additional algorithms to compute the modulator value Some unit tests are failing - will investigate further.

Adding vertices serially in TreewidthComputer.convertToUndirectedWithoutModulator() - adding vertices in parallel was causing errors and nondeterministic behavior

computeModulator() was returning an empty result when a cycle was present since the current treewidth was calculated to be the target treewidth for some algorithms for a simple cycle.

Adding eta computation shortcuts. A graph with a single vertex will always have an eta of 0, and a graph without cycles will have an eta of 1 for our purposes since we are not interested in graphs that do not have cycles

Extending timeout duration for graph computation Added TODO comment to look into using a SparseIntUndirectedGraph to perform copies

ModulatorComputerTest.ModulatorComputationTests.testCycleGraphModulator() was not generated correctly by Perplexity AI when it was generated. The graph generated is a cycle and will have a treewidth of 2.

Initial commit of OptimalKComputer and test classes

Adding no-arg solve() method to DirectedFeedbackVertexSetSolver and using the KosarajuStrongConnectivityInspector strongly connected count result as the value for k since it is a lower bound for k for a graph.

Added 1000 and 1500 node graph sizes to sparse graph benchmark test since codebases can get large but are often sparse

First pass of providing prioritized edges to remove. Will likely need to revisit how the edges are prioritized.

Updated algorithm that sorts edges that need to be removed and added unit test to confirm desired behavior

- Reordered edge disharmony removal columns - Moved edge disharmony rendering methods next to the renderEdgeDisharmonies() method

No longer computing FAS in parallel since it typically leads to a ConcurrentModificationException, even in large codebases.

…tyParallel()

…exSet()

Returning a CodebaseGraphDTO containing a classReferencesGraph instead of only a classReferencesGraph

Adding className Field to ScmLogInfo to allow mapping of class name to a file in Git. May be able to remove this in the future.

…each commit. Now caching the sum of all committed files in the repository at each commit. However, it does not appear that this provides a significant gain in speed.

Adding className Field to ScmLogInfo to allow mapping of class name to a file in Git. May be able to remove this in the future.

Returning a CodebaseGraphDTO containing a classReferencesGraph instead of only a classReferencesGraph

Calculating the priority order of edges / relationships that need to be removed to remove all cycles from the codebase

Adding HashMap as placeholder constructor value for CostBenefitCalculator since cycles are not currently being evaluated in the CsvReport or JsonReportExecutor

- Capturing edge target information to improve the ranking of relationships that need to be removed - Displaying only the largest cycle to keep page load time to a minimum while also allowing users to investigate the largest cycle.

- Making classes that need to be removed in bold instead of having a separate column indicating that it should be removed - Added spacing - Removed redundant text - Removed dead code

jimbethancourt added 30 commits July 15, 2025 20:31

Extracted EdgeRemovalCalculator to a separate class

b57c9f3

Initial commit of code as-is from Perplexity output

b6d21c1

- Fixed minor issue with lambda in DirectedFeedbackVertexSetSolver

145c9fc

- Applied Spotless

#152 Initial commit of SuperTypeToken

8ff63a2

Initial commit of SuperTypeToken. SuperTypeToken will make it possible to reify graphs in feedback algorithms when needed.

#152 Improved ergonomics

d1e80d0

#152 Now using SuperTypeToken

0853746

Now using SuperTypeToken in DirectedFeedbackVertexSetSolver and MinimumFeedbackArcSetSolver when creating new graphs in the algorithms

#152 Fixed some unit tests

12d635e

#152 Fixed more unit tests

fb58c52

No longer using parallel streams to create vertices in createRandomGraph() to ensure that all vertices are created.

#152 All MinimumFeedbackArcSetSolverTest unit tests now passing

a5a135f

All MinimumFeedbackArcSetSolverTest unit tests are now passing or are disabled with good reason

#152 No longer adding vertices in parallel

18f66b7

No longer adding vertices in parallel. Adding vertices in parallel consistently led to ArrayOutOfBounds exceptions

#152 Returning boolean instead of asserting within graph cycle checke…

56a3c49

…r method

#152 Initial commit of eta (η) and k computation

a631248

Initial commit of eta (η) and k computation. Some unit tests are failing - will investigate further.

#152 Initial commit of ModulatorComputer

33a536d

Initial commit of ModulatorComputer - includes additional algorithms to compute the modulator value Some unit tests are failing - will investigate further.

#152 Adding verticex serially

7635f56

Adding vertices serially in TreewidthComputer.convertToUndirectedWithoutModulator() - adding vertices in parallel was causing errors and nondeterministic behavior

#152 Corrected testMultipleCycles()

57c0b77

#152 Only returning modulator calculation result(s) if present

642bba4

computeModulator() was returning an empty result when a cycle was present since the current treewidth was calculated to be the target treewidth for some algorithms for a simple cycle.

Applied Spotless

897f6fe

#152 Adding eta computation shortcuts

5d6bea3

Adding eta computation shortcuts. A graph with a single vertex will always have an eta of 0, and a graph without cycles will have an eta of 1 for our purposes since we are not interested in graphs that do not have cycles

#152 Extending timeout duration

b694988

Extending timeout duration for graph computation Added TODO comment to look into using a SparseIntUndirectedGraph to perform copies

#152 Fixed incorrect unit test

e7afa03

ModulatorComputerTest.ModulatorComputationTests.testCycleGraphModulator() was not generated correctly by Perplexity AI when it was generated. The graph generated is a cycle and will have a treewidth of 2.

#152 Reverting testRandomGraphPerformance() back to original test

fbcc1bc

#152 Initial commit of OptimalKComputer

5a57c2e

Initial commit of OptimalKComputer and test classes

#152 Adding no-arg solve() method to DirectedFeedbackVertexSetSolver

36c7065

Adding no-arg solve() method to DirectedFeedbackVertexSetSolver and using the KosarajuStrongConnectivityInspector strongly connected count result as the value for k since it is a lower bound for k for a graph.

#152 Corrected testTreewidthModulator()

f24268e

#152 Updated TODO comment

32130d8

#152 Updated timeout duration

b1fb55b

#152 Moved OptimalKComputer classes to their own package

c7a700d

#152 Added comment indicating ModulatorComputer was generated

b9862f1

#152 Added benchmark sizes for sparse graphs

08c3fe2

Added 1000 and 1500 node graph sizes to sparse graph benchmark test since codebases can get large but are often sparse

jimbethancourt added 30 commits November 20, 2025 07:11

#152 First pass of providing prioritized edges to remove

418ccf6

First pass of providing prioritized edges to remove. Will likely need to revisit how the edges are prioritized.

Applied Spotless

cf40984

#152 Fixed unit tests

3ca2c43

Updated JaCoCo plugin version

53e8141

#152 Updated edge sorting

11b0e62

Updated algorithm that sorts edges that need to be removed and added unit test to confirm desired behavior

#152 Reordered edge removal columns

1b6963b

- Reordered edge disharmony removal columns - Moved edge disharmony rendering methods next to the renderEdgeDisharmonies() method

#152 Fixed typo

6ec8fb1

#152 Applied Spotless

ae44158

#152 Computing FAS in non-parallel mode

943a155

No longer computing FAS in parallel since it typically leads to a ConcurrentModificationException, even in large codebases.

#152 Using HashMap instead of ConcurrentHashMap.newKeySet();

6656391

#162 Memoizing result of ModulatorComputer.computeBetweennessCentrali…

82d9203

…tyParallel()

#162 Memoizing result of FeedbackVertexSetComputer.greedyFeedbackVert…

446d7df

…exSet()

Added comment

ff3a886

#152 Returning a CodebaseGraphDTO

dee5397

Returning a CodebaseGraphDTO containing a classReferencesGraph instead of only a classReferencesGraph

#152 Adding className field to ScmLogInfo

0c4ba56

Adding className Field to ScmLogInfo to allow mapping of class name to a file in Git. May be able to remove this in the future.

#152 Improved captureChangeCountByCommitTimestamp() to be more robust

241f0b4

#156 Now caching the sum of all committed files in the repository at …

1533279

…each commit. Now caching the sum of all committed files in the repository at each commit. However, it does not appear that this provides a significant gain in speed.

#152 Adding className field to ScmLogInfo

be68a0a

Adding className Field to ScmLogInfo to allow mapping of class name to a file in Git. May be able to remove this in the future.

#152 Added getClassName() and getPackageName() methods

f22a2c3

#152 Returning a CodebaseGraphDTO

b59452e

Returning a CodebaseGraphDTO containing a classReferencesGraph instead of only a classReferencesGraph

#152 Calculating priority of edges / relationships for removal

2890409

Calculating the priority order of edges / relationships that need to be removed to remove all cycles from the codebase

#152 Adding HashMap as placeholder constructor value

3ea2e87

Adding HashMap as placeholder constructor value for CostBenefitCalculator since cycles are not currently being evaluated in the CsvReport or JsonReportExecutor

#152 Capturing edge target information

d69ba85

- Capturing edge target information to improve the ranking of relationships that need to be removed - Displaying only the largest cycle to keep page load time to a minimum while also allowing users to investigate the largest cycle.

Increased timeout to 30 seconds for testLargeRandomGraphs()

9e0cc63

Applied Spotless

d3bce6b

#152 Corrected improved getRankedChangeProneness algorithm

e73fb94

#152 Updated sorting test to match updated sorting algorithm

5af6c04

#152 Handling null scmLogInfos better

b41981c

#152 Minor UI Enhancements in SimpleHtmlReport

668d68a

- Making classes that need to be removed in bold instead of having a separate column indicating that it should be removed - Added spacing - Removed redundant text - Removed dead code

#152 Renamed DSM module to graph-algorithms

fb72d4e

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

#152 Add minimum feedback set algorithms #165

#152 Add minimum feedback set algorithms #165

Uh oh!

jimbethancourt commented Dec 18, 2025 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

#152 Add minimum feedback set algorithms #165

Are you sure you want to change the base?

#152 Add minimum feedback set algorithms #165

Uh oh!

Conversation

jimbethancourt commented Dec 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

jimbethancourt commented Dec 18, 2025 •

edited

Loading