NaNs with interactive plot

I would like to visualize a single dataframe column / series of floats with no NaNs using interactive plotting, but the plot shows overlapping NaN values (screenshot). I am using the latest build off the master branch. And here is my code:

# cluster subset is the dataframe with no NaNs
umap.plot.output_notebook()
mapper = umap.UMAP().fit(cluster_subset)
p = umap.plot.interactive(mapper, hover_data=cluster_subset, point_size=3, theme='blue')

I am showing this in streamlit but the same thing happens in a jupyter notebook. Let me know if you have any idea what I’m doing wrong!

Thanks so much, Scott

[cchandler]

This morning I had a bunch of NaNs in my hover data despite no NaNs in the training frame. In my case it was because I had a mismatch between the indices I was providing to the ‘subset’ argument and the indices of the actual dataframe. Mine was due to selecting a subset of rows (df[df[‘some_column’] == ‘asdf’] eg idx values 3,4,9,11,12,13,15), doing a fit on those, and then accidentally providing linear, sequential numbers to subset (0,1,2,3,4). To fix I copied the DF and did a reset_index() and it fixed everything.

[lmcinnes]

This will be because some points are ending up completely disconnected from everything. The key thing to check is the metric and possibly setting disconnection_distance to something on the case of using cosine, hellinger, jaccard, or other bounded metrics.