Anthony Blaom, PhD

Maybe I need a little more context (if you still have it :wink:) to see why the error is repeating.

Maybe this isn't so bad under Julia 1.10???

@EssamWisam Can you reproduce?

Also (after marking previous as broken): ```julia Tomek Undersampler | 3 3 0.6s Progress: 100%|███████████████████████████████████████████████████████████████████████████████████| Time: 0:00:00 class: 0 cluster undersampler: Test Failed at /Users/anthony/MLJ/Imbalance/test/undersampling/cluster_undersample.jl:38 Expression: x_n ≈ X_c0_under[:, 2]...

And another: ```julia Tomek Undersampler | 3 3 0.7s Progress: 100%|███████████████████████████████████████████████████████████████████████████████████| Time: 0:00:00 class: 0 cluster undersampler: Test Failed at /Users/anthony/MLJ/Imbalance/test/undersampling/cluster_undersample.jl:53 Expression: sum((X_under[y_under .== 2, :])') ≈ sum((kmeans(X_c0, n; maxiter...

This may not be the issue, but I see there is`MersenneTwister` in testing. One should only use `StableRNG` from the StableRNGs.jl package for testing, as behaviour of `MersenneTwister` and other...

I just found that tests pass on v"1.10.10". So might be worth checking out the RNG issue first.

Here's a telling code comment: https://github.com/JuliaAI/Imbalance.jl/blob/82d7e8bc7681eea0b5c2bcc77f136bbfe71c0b66/test/undersampling/cluster_undersample.jl#L10

Yes. My best guess is that this is an RNG issue after all. We should replace all RNG's in testing with `StableRNGs.StableRNG(seed)` because MersenneTwister, Xoshiro, etc can change (do change!)...

@pebeto Do you have the possiblity to finish this review?