Bad argument for scipy.sparse.coo_matrix

[MaximeBeauchene]

I want to use umap.UMAP().fit_transform(X) but i got an error :
ValueError: negative column index found
from the scipy function scipy.sparse.coo_matrix

When i have investigated, i've found in umap_.py, in the function fuzzy_simplicial_set(), the variable tmp_indices contains values < 0 (-1), but scipy.sparse.coo_matrix need values >= 0 !

Thx for answering me.

[lmcinnes]

Hi,

Thanks for tracking down the details. In summary the issue is the following: umap uses an approximate k-neighbor algorithm to find the n_neighbors points closest to each point. Apparently that algorithm has failed to find the full n_neighbors points (with the remaining neighbors set to a flag value of -1). As you note, this breaks a lot of stuff downstream in the code. First of all some better error checking so that a more sensible error message is displayed would be useful, so I'll try to get that done ASAP. After that it is a matter of tracking down why the approx knn algorithm is failing, which may take longer to sort out.

[lmcinnes]

I believe I have resolved the error, but would appreciate it if you could check if that is the case using the current master. Let me know if it works now, thanks!

[MaximeBeauchene]

I have cloned your git :
git clone https://github.com/lmcinnes/umap.git

After i have done : python3 setup.py install
And i still have the same issue !

[MaximeBeauchene]

The git log say the last commit is :

commit da6fe79
Author: Leland McInnes leland.mcinnes@gmail.com
Date: Mon Nov 13 11:46:01 2017 -0500

Match new calling convention better

5 hours before you answered me, have you push the fix ?

[lmcinnes]

I did indeed push the fix. I have found that I sometimes need to purge the previous version from my site-packages directory prior to a re-install before it takes. Could that be an issue?

[MaximeBeauchene]

Thanks a lot !

It is working great.

Good job.