Talk:Moving average/Archives/2019

Question on Application to measuring computer performance

I would like to know for which application these weights are suitable? Formulas look like they're intended for an arbitrary measurement times $t_{n}$. But say we have a time constant of 1 minutes (W = 1), and we get new measurements values at t = 0, t = 1000, t = 1001. Then we will get very different weights for the last two measurements, weight close to 1 for t = 1000, and close to 0 for t = 1001. Which doesn't make a lot of sense for the applications I can imagine; measurements close in time in relation to the time constant ought to get roughly equal weights.

2620:0:1043:12:84C:D06D:A4B7:59D4 (talk) 14:55, 22 November 2019 (UTC)

Seems to me that the page talks about an average over a value series, which is appropriate in a time-series use case when the samples arrive on a regular basis. When the samples themselves have additional information such as time of measurement, and the task is to estimate the average value around the "now" time rather than the average value in the current point in the series, I suspect the math will change significantly, but I'm not sure how. --Alvestrand (talk) 07:27, 25 November 2019 (UTC)