2
$\begingroup$

I'm trying to develop a change detection model that uses sliding windows. Given a time series with some features I've a sliding widows that analyses that time period and compares with a successive time window. If the distribution of data has changed than it is notified. Anyway, I don't have enough real data about my problem, so I need to generate some artificial time series but with different distributions, no matter what distributions, the important thing is that they are different, so that I can test when my model detects change in the data. How can I generate those data?

$\endgroup$
1
$\begingroup$

If I understand you correctly, I think I have a simple way and you do not need to overthink it.

  1. Take a Gaussian (Normal) distribution random number generator, it could even be just downloading a set from https://www.random.org/gaussian-distributions/
  2. Multiply by the desired standard deviation, add the desired mean.
  3. From a certain timestamp, change the standard deviation and mean to a new value to simulate the distribution change. This is the same as concatenating smaller sets with different mean and standard deviation.
  4. See if/when your model detects it.
$\endgroup$
  • $\begingroup$ Thank you very much. Anyway, do you think that I should consider that because this should be a time series is not so "correct" to create a series with only values taken from a normal distribution? Should be values of a series correlated among themselves? $\endgroup$ – exrezzo Jan 17 '20 at 10:41
  • $\begingroup$ feel free to modify my proposal by adding the previous value, you probably have at least some idea on what your time series should look like.. $\endgroup$ – Pieter21 Jan 17 '20 at 12:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.