Question about using DBA to create a average curve from several input curves

Hi,

I have a question about using DBA to get the average time series of a data set consisting of several measurements. I have collected the knee joint angles during gait for several subjects. Now I want to create the average knee flexion from this data. During my search on how to do this I found out about using DTW for averaging and thought this might a good way since in takes into consideration the slightly different curve shapes in constrast to a simple mean over all curves.

However the results I get from using DBA look very strange to me. For example using softdtw_barycenter with gamma=1, max_iter=50 and tol=1e-3 I get the following result, where the black line is the result and the colored lines are the individual input curves:

Here I dont understand where the spikes come from. This also happens when I try different parameters or other algorithms like dtw_barycenter_averaging.

On contrast when I use the euclidean_barycenter the output looks more like I would expect:

So now my question is: Is the result due to my data or did I select the wrong parameters? Or is the result supposed to look this way and DBA is simply not the right way to get a physiologically reasonable average knee flexion?

Many thanks in advance!

Best regards,
michael

The report concerns softdtw_barycenter and dtw_barycenter_averaging, but it provides no runnable inputs, code, or named repository files. First request a minimal reproducible example with the input curves and parameters, then compare both barycenter methods with euclidean_barycenter. Done means determining whether the spikes are expected for these inputs or reveal a reproducible implementation or documentation problem.