Leland McInnes
Leland McInnes
I tried pushing a new version of umap to pip that should address this issue. Hopefully that may help. On Tue, Oct 9, 2018 at 8:15 AM Jeremy Manning wrote:...
I thought, in fact, that this is something that had been fixed. The latest version in the repository should, I think, handle this correctly. I just haven't made a release...
Hmm, that's less good then :-( There is, at least, theoretically code that guards around this now. I'll have a look soon.
The short answer is that the sample_weights should effectively simply be all ones unless otherwise specified. In the newer code there is actually a bifurcation and a separate version of...
Hmm, that's more disconcerting. I have been doing a lot of messing around with it a month ago and may have subtly broken something. I'll try to get some time...
Right now I have many other pressing things. I am unlikely to get to this until November at the earliest. As a workaround in the meantime you can force the...
It is in a bit of limbo right now as I have had to step away from this library to work on other topics.
For large datasets I think this is not unreasonable. In principle I would think that hierarchical clustering with single linkage could be made almost as fast, but you would be...
So it looks like we can just use a positional argument and support both versions. I'll see if I can get this done soon.
You can provide a custom metric if you wish as long as it can be jit compiled by numba. Just specify the function as the metric instead of a metric...