If you have a lot “spikey” histograms, a kernel density estimate to smooth it out is really nice. When I’m using matplotlib I just add kde=True and it’s done. As far as I can tell, there’s no simple way of creating this on any visual in Quicksight. (I would be so happy if this could get on a roadmap somewhere…)
My options at the moment:
I can create a KDE in python with matplotlib, store the result in s3 and use a URL viewer. Downside: can’t do any filtering this way, unless I pre-compute a graph for every slice I’ll every plan to use. (And even then, how would I get the right image displayed in quicksight?)
I can create a view in postgresql. I tried doing this initially, but the SQL to create a kernel density estimate is pretty hideous and doesn’t compute very quickly.
Any other ideas or suggestions of how I can make it happen?