Improve initial population diversity in IsoFuel algorithm

[jnakaigen]

Description

I have been reviewing the IsoFuelPopulation class in algorithms/genetic/population.py and noticed a potential issue with the fallback strategy for initial population generation.

Currently, in IsoFuelPopulation.generate, if the patcher returns fewer routes than n_samples, the code fills the remaining slots by copying the last available route (lines 163-165):

for j in range(i + 1, n_samples):
    X[j, 0] = np.copy(X[j - 1, 0])

Impact:
If IsoFuel generates only a few valid routes (e.g., 3 out of 50), the remaining 47 individuals become identical clones. This severely limits genetic diversity in the first generation, potentially causing premature convergence or stagnation.

Proposed Solution:
I propose adding a perturbation step to the fallback mechanism. Instead of exact copying, we could apply a small random mutation to the coordinates of the copied routes to ensure the initial population maintains diversity.

I am happy to draft a PR for this if this approach aligns with the project's goals.

[AGmitmanipal]

Hi,

Thanks for the detailed explanation. This makes a lot of sense.

I went through "IsoFuelPopulation.generate" and agree that copying the last route can lead to a highly homogeneous initial population, especially when the patcher produces few valid routes.

Before starting on a PR, I wanted to confirm:

• Would a light perturbation/mutation-based fallback align with the intended behavior of IsoFuel?
• Is there a preferred mutation strategy or existing utility for small coordinate noise that I should reuse?

I’d be happy to draft a PR once I get your confirmation.

Thanks!

[jnakaigen]

Hi @AGmitmanipal, thanks for validating the finding!

I have already drafted the fix locally (implementing a Gaussian perturbation fallback) and I am just verifying it against mutation.py to ensure I'm reusing existing utilities.

I will push the PR shortly.

[codehashira99]

Hi, I’m interested in contributing to 52°North.
Could someone please share the Slack invite link for 52°North?
Thank you!

[azad2003-ai]

Description

I have been reviewing the IsoFuelPopulation class in algorithms/genetic/population.py and noticed a potential issue with the fallback strategy for initial population generation.

Currently, in IsoFuelPopulation.generate, if the patcher returns fewer routes than n_samples, the code fills the remaining slots by copying the last available route (lines 163-165):

for j in range(i + 1, n_samples):
X[j, 0] = np.copy(X[j - 1, 0])
Impact: If IsoFuel generates only a few valid routes (e.g., 3 out of 50), the remaining 47 individuals become identical clones. This severely limits genetic diversity in the first generation, potentially causing premature convergence or stagnation.

Proposed Solution: I propose adding a perturbation step to the fallback mechanism. Instead of exact copying, we could apply a small random mutation to the coordinates of the copied routes to ensure the initial population maintains diversity.

I am happy to draft a PR for this if this approach aligns with the project's goals.