Add white noise to series
Opened this issue · 2 comments
Imputed values look totally smoothed out for a lot of algorithms and because of that those values do not respect the statistic of the rest of the signal.
This will for example influence standard deviation of the series...
Interesting solutions to this could be an option for adding "white noise" with roughly the same standard deviation as the rest of the time series.
Offering Multiple Imputations could be another option.
Downside here is the MI concept is often misunderstood by users and this solution would harm the simplicity of use.
Or other colored noise possibly more appropriate for timeseries data? See https://github.com/japilo/colorednoise
Thanks a lot @Aariq ! That is pretty interesting, thanks for sharing the link!
From the github:
'Many populations that change over time are temporally autocorrelated, which means that the random noise in each timestep is correlated to that of the previous timestep.'.
This is definitely true for a lot of time series. Especially the non-stationary ones. Even in stationary time series the random noise in each timestep can still be correlated to that of the previous timestep.
That is probably, why it is already on the To-Do Iist for some time ... not as easy as it looks. I definitely have to dig deeper into the topic before adding something.