Covers vs. Bins

[tomtomato]

Can you please explain the distinction between 1) the number of covers (for each of the "n" lenses), and 2) the number of Bins (in the clustering section)?

My understanding is that the covers on the lenses jointly are used to define the overlapping bins for clustering the inverse image of the data. So I'm not sure what the bins parameter does in the clustering section. Thanks!

[rosinality]

Number of cover specifies number of overlapping interval in low dimensional embedded space. Cluster binning is, as specified in G. Singh, F. Memoli, G. Carlsson (2007), used for determining number of clusters in the cover. After dendrogram is constructed, histogram of distances between data points is computed. So binning specifies number of histogram bins. After that, dendrogram is split at the point that last "gap" of histogram was occur. That is, for example if histogram was some like this:

• 0-1 : 3
• 1-2 : 0
• 2-3 : 4
• 3-4 : 0
• 4-5 : 2

Then dendrogram will split at 3-4.

So binning determines the number of clusters in somewhat indirect way.

[tomtomato]

Thank you - that's a very clear description.