min_dist not updating with Python backend

[lgaborini]

Hello, I found out this pretty strange bug.

In short, when I use the umap-learn backend, any setting of min_dist is not respected.
The Python backend complains if it's higher than spread, the arguments change, min_dist is in
config$umap_learn_args, but the UMAP coordinates do not change.

Other settings (e.g. n_neighbors) are good.

When the same code is called directly from a Python session, min_dist is correctly applied.

Repro and more comments in this repository:
https://github.com/lgaborini/umap-bug/blob/master/umap_compare_iris.md

Source (R markdown + reticulate)

Thank you!

[tkonopka]

Thanks for the bug report and the detailed examples. Can confirm that I can reproduce the effect. WIll investigate.

[tkonopka]

Thanks again for spotting this bug. I just committed an update; it fixes the issue as far as I can tell. Would you be able to try it out too? Many thanks.

Here's more background. min_dist is used together with spread to tune how close together items should appear in the final embedding. An alternative way to achieve the same tuning is via parameters a and b. The latter pair actually take precedence. By default, a and b are None in umap-learn, so changing min_dist (and/or spread) is an effective way of altering the embedding. But if a and b are set, values assigned to min_dist or spread become irrelevant.

The bug in the R package arose because NAs for a and b in the configuration object were mistakenly translated into python number-like values. Those number-like values for a and b made min_dist inactive; calculations with different min_dist produced equivalent coordinates. The fix avoids sending NA values to umap-learn, thus avoids setting a and b, and thus enables min_dist to take effect.

[lgaborini]

Looking great also in my code! Thank you!