Possible Typo on the get_cluster_label_vector (line 100)

[zackaryleady]

I am trying to use this incredible new library you are working on. And following your example in the arxiv notebook I received this error:

It appears that an argument maybe missing (possibly a difference between HDBSCAN and the fast_hdbscan libraries?

From fast_hdbscan cluster_trees.py it looks like the function needs 4 inputs, but only 3 are provided. @numba.njit() def get_cluster_label_vector( tree, clusters, cluster_selection_epsilon, n_samples, ):

If you could provide a reasonable value for cluster_selection_epsilon and/or n_samples.

My Code if it helps:

[lmcinnes]

It is an internal method of fast_hdbscan that got updated; I believe toponymy has updates, but potentially they haven't been pushed to github yet. This library is under pretty active development right now, and for the moment you should expect random breakages to happen. Hopefully we'll get it to more stable state soon and make an actual release to PyPI that people can rely on a little more.

[zackaryleady]

Hello,

I just wanted to say that the latest fix by John resolved the issue. I did need to wrap my llm in the LlamaCPPWrapper() because otherwise there is an issue with the llm_instruction() not being found. Once I used the LlamaCPPWrapper() I also had to remove the encode("utf-8") in my own code, because the Wrapper does this internally. If you think it would be useful I could write down some helpful hints on all the debugging I had to do to get it working for future users, just let me know.

~Zack

[lmcinnes]

We are moving away from the LLamaCPP because the online services are easier and fairly popular, but if you have fixes that makes the LlamaCPPWrapper work better that would be greatly appreciated.