Matching a target time series to other time series using dynamic time warping

Hello,

I have one data matrix A of size = n by m , where n is the number of observations and m is the number of variables, (f.ex. 1000 by 20)

I also have a target time series x of size n by 1 (f.ex 1000 by 1)

I would like to use dynamic time warping to find a cluster in A that has the smallest distance to x relative to other variables in A.

Essentially, find the variables in A that are most similar to the target variable x.

How can I do this?

Kind regards

Knut

[Eamonn]

Are you sure you need DTW?

If Eucldean distance would work, the MASS is unbelievably fast (and easy) (matlab code here [a])

If you do need to use DTW, you probably need to set the warping window width , see https://www.cs.unm.edu/~mueen/DTW.pdf

[a] https://www.cs.unm.edu/~mueen/FastestSimilaritySearch.html

[Knut Skaug]

Thank you for the suggestion - I will try it out

Answer 1

[Star Strider]

See if the Signal Processing Toolbox dtw function (introduced in R2016a), or related functions linked to in that page, will do what you want. (I have used findsignal, however not dtw, so I have no specific experience with it.)