-
Notifications
You must be signed in to change notification settings - Fork 7
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Handle unevenly sampled time series #32
Comments
Oof, i've thought about this extensively but I haven't written anything here. Sorry, it would have saved you a lot of effort... From my eyes, the plan is to create a package TimeseriesViewers.jl that defines a
The WindowViewers then produce views over timeseries with a given window an stride, in real time units, or integer units for convenience if the timeseris is evenly sampled. The main takeaway here is that TransitionIndicators.jl will not care at all whether the timeseries is evenly sampled or not, because it gets in a |
it is unfortunate that you went though a lot of time to write your code, but I really think the best way forwards is to just rely on Interpolations.jl for creating windows of an unevenly sampled timeseries... |
From a design perspective I agree with you. This would be the easiest. However, one might want to avoid interpolation in some cases, as it biases e.g. the AR1-regression coefficient and the power density spectrum. Going the way you propose means closing that option, but it might be a choice we want to make. Let's talk about it live at some point. |
Not sure what you mean here. Avoiding interpolation when? The user can choose to not construct the interpolation object, and rather just given in an abstract vector directly, in which case no interpolation would/coould happen. |
I am referring to this sentence. I understood that you (always) want to access unevenly-spaced time series by interpolations. But maybe that was not your point. Let's talk about it live to make sure we understand each other. |
Is your feature request related to a problem? Please describe.
Empirical data are often unevenly sampled. In the state of the art, an interpolation is often performed but this might bias the result.
In
TransitionIndicators.jl
we want to offer both options: (1) the user can interpolate their data and pass it or (2) directly pass their unevenly sampled time series, in which case no interpolation happens and all the difficulties are handled internally. (Option 2) might have reduced features compared to (Option 1) but introduces no error.We won't focus too much on (Option 2) for now but defining the basics for the API to handle it should be a step in near future.
Describe the solution you'd like
Here goes an idea of solution. It does not rely on defining an iterator (maybe a downside?) but has a quite sparse syntax:
Describe alternatives you've considered
Additional context
One open question is whether the window width on unevenly time series should be defined by the number of points or by a time span. I think the latter one is more rigorous.
The text was updated successfully, but these errors were encountered: