Hi,
I would like to do kDBA, but using a custom metric for computing the DTW alignments (not available in either scikit or scipy).
Now, dtw_variants has the dtw_path_from_metric function, so there it is always possible to compute dtw alignments for any pair of timeseries, passing the metric either as a functional, or by passing a pre-computed distance matrix (which for my purposes is not super-feasible).
Now, the problem is that the dba module only ever uses dtw_path, and so is always, at least implicitly, using euclidean distance. Is there a straigtforward way of passing custom metrics to TimeSeriesKMeans such that it computes the dtw alignments using a custom metric? It probably isn't too hard to change all calls to dtw_path to dtw_path_from_metric, conditional on a metric being provided. Just checking if there is an easier way.
Thanks,
Hi,
I would like to do kDBA, but using a custom metric for computing the DTW alignments (not available in either
scikitorscipy).Now,
dtw_variantshas thedtw_path_from_metricfunction, so there it is always possible to compute dtw alignments for any pair of timeseries, passing the metric either as a functional, or by passing a pre-computed distance matrix (which for my purposes is not super-feasible).Now, the problem is that the
dbamodule only ever usesdtw_path, and so is always, at least implicitly, using euclidean distance. Is there a straigtforward way of passing custom metrics to TimeSeriesKMeans such that it computes the dtw alignments using a custom metric? It probably isn't too hard to change all calls todtw_pathtodtw_path_from_metric, conditional on a metric being provided. Just checking if there is an easier way.Thanks,