I've been trying to cluster time series of shape (1, 400), so 1 row and 400 columns which correspond to 400 timesteps. My train set is of size (1000, 400) so 1000 time series. I have calculated a distance matrix with DTW and used it with scipy fcluster :
I'm quite satisfied with the result. I wanted to know how I can use this trained model to cluster new time series? I have 23 clusters.
Should I add the new time series with the old and calculate a new distance matrix? How do you save such a model?