Genetic Algorithm: Crossover skipped for small routes may reduce diversity

[Gnananjali]

While experimenting with constrained configurations (e.g., small routing steps), I observed that crossover is skipped when route length is small:

if p1.shape[0] <= 5 or p2.shape[0] <= 5:
    return p1, p2

This effectively disables crossover, reducing genetic diversity and potentially increasing the likelihood of no feasible solution being found.

This behavior may impact GA performance in constrained scenarios.

Possible considerations:

• Alternative crossover strategy for small routes
• Fallback mutation or diversification mechanisms
• Validation or warnings for such configurations

I have added tests to validate this behavior in a separate PR.

[arybhatt4533]

Hi Martin @kdemmich

Following up on my previous message, I have officially submitted my proposal for the WRT Test Suite modernization project on the GSoC portal.

I would also like to ask where the team prefers to hold detailed technical discussions regarding the project and my proposal. Should I continue the conversation here on GitHub, or is there a specific Zulip/Slack channel or community forum I should join to better align with the team's workflow?

Looking forward to your guidance!

Best regards,
Arybhatt

[arybhatt4533]

Hi @Gnananjali, thanks for highlighting this. The current threshold of <= 5 indeed creates a bottleneck for diversity in constrained routing scenarios.

As part of my GSoC proposal regarding the WRT Test Suite modernization, I am particularly interested in how we validate these edge cases. I’d love to contribute a fix that introduces an alternative crossover strategy or a fallback mutation for small shapes, ensuring diversity isn't compromised.

I can also write a dedicated Pytest case to reproduce this behavior and verify the fix. Should I start by drafting a PR with a reproduction test?

[Gnananjali]

Hi @arybhatt4533, thanks for the insight!

You're right that skipping crossover for short routes can reduce diversity in constrained scenarios.

In this case, the main goal was to prevent the runtime crash caused by invalid bounds in the crossover step. I agree that introducing an alternative crossover strategy or fallback mutation would be a valuable improvement.

Adding dedicated tests for this behavior also sounds like a great idea. Happy to collaborate or help test further enhancements!

[arybhatt4533]

Hi @Gnananjali and @kdemmich,

While waiting for the discussion channel details, I've decided to move forward with the Local Environment Setup.

I am currently drafting a Pytest script to formally reproduce the diversity bottleneck in Issue #189. My goal is to demonstrate how the shape[0] <= 5 skip leads to identical offspring. Once I have the reproduction test ready, I will share the results here as a Draft PR.

Looking forward to showing some initial results soon!