0

Is it sound/allowed to use entropy as a measure of (non-) uniformity of a data stream?

E.g. I calculate the Shannon entropy with the standard formula based on various measures in the data stream. The measures are count values whose probabilities sum up to 1. I do the calculation continuously every 1 sec, so I get a new entropy value every second. Then I monitor the average and standard deviation of the entropy values over a sliding window, let's say the past 10 seconds. If the stddev increases or gets larger than certain threshold then I assume the stream got noiser/bumpier. Is it ok to use entropy in this context?

Andreas
  • 579
  • 1
  • 5
  • 19
  • I would argue that if the mean increases your stream has more noise. The standard deviation of the entropy says more about how stable your calculation is. It could be used to create reasonable cutoffs (i.e. z-score) before you take action. – muratoa Sep 30 '13 at 19:55
  • Yes, of course. If the mean increases the noise increases. And the less the stddev the more stable the stream. So I can use this interpretation of entropy monitoring for my stream analysis, I'd say. – Andreas Oct 01 '13 at 14:52

0 Answers0