ClasPy for short duration time series

[DanielaFe7-personal]

Hello!
Thanks for your work
I was wondering if the library is right for "short" time series.
Our data set is about spectral indices derivated from satellite data.
We have for intance, NDVI time series for 38 years as seasonal data (4 points per year).
This is only 152 points composing the time series.
Thus when running the algorithm for the segmentation No changepoint is encountered.
Data is already interpolated as a numpy array.
But across the whole dataset, no changepoint is detected.
I was wondering what parameter could be modify to allows the segmentation.
Thank you!.

[ermshaua]

Hi,

thanks for reaching out! ClaSP was mainly designed to segment medium to long time series, e.g. with >1k data points. You can, however, reconfigure it to be used for short time series.

Try setting the window size to 3 or 5 data points manually. Also, disable the automatic detection of the number of change points and try setting it manually. If you are using NDVI data, consider changing the distance function to the normal Euclidean distance.

You can find a detailed tutorial on how to set these hyperparameters in this notebook: https://github.com/ermshaua/claspy/blob/main/claspy/notebooks/clasp_configuration.ipynb

Let me know if this helps.

Best
Arik