scikit-learn-contrib / hdbscan

A high performance implementation of HDBSCAN clustering.

http://hdbscan.readthedocs.io/en/latest/

BSD 3-Clause "New" or "Revised" License

2.81k stars 507 forks source link

Create a patch release with updated Cython dependencies #646

Closed vyasr closed 3 months ago

vyasr commented 4 months ago

644 updated the build dependency of hdbscan to support using Cython 3.0. However, this update has not yet been released. Furthermore, there are still issues with using hdbscan in monolithic build/run environments because Cython<3 is still included in the runtime dependency list. Rather than relaxing that pin, though, it could simply be removed because Cython is not required at runtime at all, only at build time. The necessary change is available in #624.

Would it be possible to merge #624 and then cut a release? The current state of the latest release (sans #624 and #644) is blocking cuml from being able to use the latest version of hdbscan (see https://github.com/rapidsai/cuml/pull/5906), so getting things updated would be much appreciated. Thank you!

CC @cjnolet @divyegala @trxcllnt

jakirkham commented 4 months ago

It's also worth noting there is a fairly loud warning coming from the test suite. Not sure if we are seeing this anywhere else. Might be nice to fix

https://github.com/scikit-learn-contrib/hdbscan/issues/647

cc @seberg

seberg commented 4 months ago

Yeah, I had ignored it to not to do much at once and it wasn't new (and locally it isn't as noise, since it originates always from roughly the same place).

Seems to just be:

diff --git a/hdbscan/_hdbscan_tree.pyx b/hdbscan/_hdbscan_tree.pyx
index aeb4051..420724f 100644
--- a/hdbscan/_hdbscan_tree.pyx
+++ b/hdbscan/_hdbscan_tree.pyx
@@ -623,7 +623,7 @@ cpdef list get_cluster_tree_leaves(np.ndarray cluster_tree):
 cpdef np.intp_t traverse_upwards(np.ndarray cluster_tree, np.double_t cluster_selection_epsilon, np.intp_t leaf, np.intp_t allow_single_cluster):

     root = cluster_tree['parent'].min()
-    parent = cluster_tree[cluster_tree['child'] == leaf]['parent']
+    parent = cluster_tree[cluster_tree['child'] == leaf]['parent'][0]
     if parent == root:
         if allow_single_cluster:
             return parent

jakirkham commented 3 months ago

@lmcinnes is there anything we can do to help here? 🙂

Please feel free to give us tasks. We would be happy to chip in

lmcinnes commented 3 months ago

Sorry I've been away on vacation. I'll try to take a look at this soon.

jakirkham commented 3 months ago

No worries. Thanks for your help Leland! 🙏

jakirkham commented 3 months ago

Looks like this was done with 0.8.38

Should we close?