Input data is a short time series, and I want to find a scalable short series that looks like the input from a long time series, should I use the sliding window, or do you have any better suggestions?
1 Answer
$\begingroup$
$\endgroup$
1
Can you clearly divide the short sequences or are they merged together?
If they have a clear time division (ex: a day or a week), you can set a batch of a short time and learn every short sequence correctly.
Otherwise, you have to find ways to cut the short sequences correctly without overlapping each other.
In both cases, you can detect short sequences' shapes thanks to 2D CNN for instance.
answered Oct 24, 2022 at 8:10
-
$\begingroup$ Does it answer your question? If not, please let me know. $\endgroup$ Commented Oct 29, 2022 at 14:54