Sudden outlier

[Yorko]

I've quickly tested Multicore-TSNE and umap on the telecom churn dataset.
Here is the notebook. The dataset is available in the very same repository.

Looks like umap suffered from a sudden outlier which hadn't affected t-SNE.

I haven't played with umap hyperparameters but maybe the example will be useful.

[lmcinnes]

Thanks. As evidenced by issue #2 the current experimental code is not robust against some corner cases. I working to find and fix those now, and this will certainly be highly beneficial in understanding where and how I can make the code and algorithm more robust.

[lmcinnes]

You may also wish to try increasing the spread parameter to alleviate this. The default is 1.0, values more like 2.0 or higher may help to reduce the effect of such outliers. If it is a singleton outlier that may require some other changes in the code that I am working on.

[lmcinnes]

I just checked in some code that should help with this problem. See if you can pull from the latest master and rebuild and let me know if that helps. Thanks!

[sleighsoft]

Did this get resolved? Can the issue be closed?

[Yorko]

Yes, thanks!