As stated here, scikit-learn has released 1.8, which removed "force_all_finite" in favour of "ensure_all_finite". This breaks new installs of hdbscan that are on python 3.11 or above, which is what scikit-learn 1.8 is compatible with (e.g., python 3.10 will only install up to sklearn 1.7, and so this is still just a deprecation warning in those cases.
I would recommend immediately cutting a release that locks hdbscan to scikit-learn<1.8, to avoid confusion and failures in workflows.
Then it's less urgent to migrate to ensure_all_finite. I assume that not as simple as it appears, perhaps due to intentions of maintaining backward compat :)
Related:
As stated here, scikit-learn has released 1.8, which removed "force_all_finite" in favour of "ensure_all_finite". This breaks new installs of hdbscan that are on python 3.11 or above, which is what scikit-learn 1.8 is compatible with (e.g., python 3.10 will only install up to sklearn 1.7, and so this is still just a deprecation warning in those cases.
I would recommend immediately cutting a release that locks hdbscan to scikit-learn<1.8, to avoid confusion and failures in workflows.
Then it's less urgent to migrate to ensure_all_finite. I assume that not as simple as it appears, perhaps due to intentions of maintaining backward compat :)
Related: