I have the following problem: I have a time series with counted data. I now want to smooth it using a Gaussian low-pass filter. Is there a method to determine the sigma value? The window should have a size of 365. The data set contains 2191 entries.

In the present study, from which I got the idea, only the filter weights are mentioned. Is this perhaps what is meant by this? It was described with: "required filter weights can be calculated from the standardized normal distribution and its density function". If so, how should this work (I get at the end a function and not a discrete value that I can use for sigma)?

  • $\begingroup$ You seem to be asking two different questions. One is "given what I know about my data, what's the best sigma value to use for a Gaussian filter"? The other is "given some filter weights for a Gaussian filter, how do I determine its sigma value, and how do I generate my own filter"? So -- which question are you asking? If it's the first one, we need to know more about your data. If it's the second one, post the weights, or if the table is large post the center weight with its index, and the weights at about 1/3 down from center (with indexes) and about 2/3 down from center. $\endgroup$
    – TimWescott
    Commented Apr 8, 2021 at 16:13
  • $\begingroup$ Oh, and because StackExchange is quirky, please edit your question with any clarifications. $\endgroup$
    – TimWescott
    Commented Apr 8, 2021 at 16:14


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.