0

I performed some computations on a time-series, and when I checked the probability distribution of a result I obtained this odd histogram:

enter image description here

I've never seen gothic spires like this. What kind of distributions is this? Are the spires the result of some elaboration artifact?

EDIT: This is the result in time-series form.

enter image description here

Fabio Capezzuoli
  • 125
  • It looks like the superposition of two normals with the same mean (one having significantly lower variance). there is some visible skew, but without knowing more it's hard to tell if that is an accident or not. – lulu Sep 24 '18 at 12:03
  • Giving more details is hard because it's the result of quite a few steps of computation. There is some periodicity in the result when plotted as time-series, could that be the reason of the odd distribution? – Fabio Capezzuoli Sep 24 '18 at 12:13
  • Periodicity? Well, absent context it's hard to guess what you are looking at there. As I say, I don't think this looks particularly odd. It just looks like there are two phenomena going on at once. they have the same (or similar) means which ought to be a strong clue. The one with higher variance appears to be discrete, hence the spikes. Still, dangerous to read too much into a single plot. – lulu Sep 24 '18 at 12:16
  • I'd try to fit two normals to the data (or skew-normals if the apparent skew is significant). See if you can makes sense out of the two sources of randomness you get that way. – lulu Sep 24 '18 at 12:18
  • Perhaps the output is quantised at a slightly different interval from the bin width. Once in a while, two quantum values are in the same bin, which gets double the frequency. – Empy2 Sep 24 '18 at 13:18
  • ... or three quantum values instead of two, to give a 50% boost. – Empy2 Sep 24 '18 at 13:29
  • It must be what @Empy2 said, because setting manually the number of bins to 100 gives a distribution without strange features. – Fabio Capezzuoli Sep 24 '18 at 16:21

0 Answers0