Skip to content

Clarify WindowSize parameter in ForecastBySsa #5959

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

Open
TomasMartincik opened this issue Oct 6, 2021 · 1 comment
Open

Clarify WindowSize parameter in ForecastBySsa #5959

TomasMartincik opened this issue Oct 6, 2021 · 1 comment
Labels
documentation Related to documentation of ML.NET Time Series

Comments

@TomasMartincik
Copy link

Is your feature request related to a problem? Please describe.
The recommended value for the WindowSize parameter of the ForecastBySsa algorithm is unclear. Available samples, as well as the example in official documentation, seem to suggest that the recommended value for WindowSize is equal to the length of the cycle present in the timeseries data. Simple tests reveal that this is not the case, and a good recommended value seems to be WindowSize=cycleLength+1, rather than WindowSize=cycleLength. Even in the documentation example linked above, which uses data with a clear cycle of length 5, the prediction does not work well with WindowSize = 5, but works perfectly with WindowSize = 6.

Describe the solution you'd like
Please clarify if there is a recommended value for WindowSize and if so, update the documentation/samples to convey it clearly.

Describe alternatives you've considered
Checked implementation of the algorithm, publicly available samples using the algorithm, as well as the scientific paper on which the alogirthm is based - didn't get a clear asnwer.

Additional context
Documentation example of the related DetectSpikeBySsa algorithm, suggests that the recommended value for SeasonalityWindowSize=cycleLength+1.

@michaelgsharp michaelgsharp added documentation Related to documentation of ML.NET Time Series labels Oct 14, 2021
@michaelgsharp
Copy link
Contributor

I'm wondering if its related to #5891. Thanks for the heads up.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
documentation Related to documentation of ML.NET Time Series
Projects
None yet
Development

No branches or pull requests

2 participants