Should the first neighbor be skipped during when transforming new data?

[jlmelville]

In smooth_knn, after calculating rho, the sum of affinities loop starts at 1:

umap/umap/umap_.py

Line 220 in 300cbba

for j in range(1, distances.shape[1]):

This makes sense to me in the standard fit case, where neighbor 0 of point i is going to be i itself. But what about in the case when transform is called on new data? This neighbor is likely to have a non-zero distance and so will be included in the calculation of rho. Technically, should the target value be np.log2(k + 1) * bandwidth in the case of transforming new data?

I realize that this is unlikely to be very important, but I was rewriting the equivalent part of uwot and wanted to make sure I hadn't introduced an error.