mblondel / soft-dtw

Python implementation of soft-DTW.
BSD 2-Clause "Simplified" License
550 stars 98 forks source link

Add Tolerance to S-DTW #18

Open Steenbuck opened 5 years ago

Steenbuck commented 5 years ago

Hey,

for my masterthesis on I want to implement a tolerance proposed in this paper: Time-series Averaging Using Constrained Dynamic Time Warping with Tolerance The aim is: "To better represent a set of signals, we added to its modelling the tolerance that depicts admissible values around the average signal. This tolerance is computed for each time step of the average signal. It corresponds to the standard deviation of the sets of signals aligned on each point of the average time-series. The tolerance of the CDBA at time step k is thus computed as the standard deviation of the points aligned on this time step k of the average time-series." The pseudo-code looks like this:

cdba-pseudocode

In order to produce a figure like this we need to add the standard deviation for the list of points which are regarded for computing each respective point of the centroid: Average signal with CDBA for two different sets of signals. The tolerance corresponds to the gray area around the average signal, computed as ±(1 × σ).

Can someone point me in the right direction where to start? Like in which function are the centroids computed and which variables I need to mess with?

Thank you very much