@Micky774 can you do a quick benchmark to compare your PR and main on float32 to be sure it does not introduce any performance regression ?

@Micky774 can you do a quick benchmark to compare your PR and main on float32 to be sure it does not introduce any performance regression ?

@jeremiedbb Benching w/

# %%
from sklearn.cluster import Birch
from sklearn.datasets import make_blobs

X, y = make_blobs(n_samples=1000)
brc = Birch(n_clusters=None)

# %%
%timeit brc.fit_transform(X)

branch: 45.1 ms ± 983 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
main: 44.5 ms ± 1.87 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)

No (visible) performance regression.

@jeremiedbb Wanted to check if you had any other feedback on this PR -- thanks :)

Thanks @Micky774 !