DTW is implemented in both Matlab and C/MEX. The C/MEX function is very fast. The algorithm we use is the standard one from wiki: http://en.wikipedia.org/wiki/Dynamic_time_warping
There is a demo in the package. To use, just call:
d=dtw(s,t,w) or d=dtw_c(s,t,w)
The function does not work for vectors of different length
Something is wrong...
should be 12
dtw_wiki = dtw(s2',s1',10)
plot(s1,'DisplayName','s1');hold on;plot(s2,'DisplayName','s2');grid;hold off;
Hi Quan, thanks for your contribution. I have a request which I am not sure it is proper. Do you mind adding the warping path to the output? It could be appreciated a lot if you could. But if it takes lots of effort, never mind. Thanks for your contribution.
Works well. Thank you very much. Especially the C/MEX one is very fast.
x = -10:0.1:10;
y = sin(x);
len = length(y);
y2 = interp1(1:len,y,linspace(1,len,2*len));
I'm very new to dynamic time warping. Does "D" indicate the similarity between two vectors?
What is a common value for the w parameter?
Just wonder, which is the corresponding paper for this code?
It is nice. It would be useful if you can add an option to force the alighment of start/end points of two signals. Or force the alignment of selected points (so it becomes interpolation in a sense).
Hi Quan, This is a great piece of work and I have made slight changes to normalize the DTW distance by its warping path for both matlab and c versions for my project. Please let me know if I can contribute that here.
Could i ask you a question? The code works perfectly when two input data have the same length. But if two input data have different length, such as [10 12 14 18 14 12 10 10 12 14 18 14 12 10] and [NaN 12 14 18 14 12 10 10 12 14 18 NaN NaN NaN], the output is NaN. Could you think how to deal with it?
Hi Quan Wang.
I have a question:
What the paper does algorithm DTW using adapt window size belong to ?
I am getting error as
'Maximum variable size allowed by the program is exceeded.'
for below line.
D=zeros(ns+1,nt+1)+Inf; % cache matrix
The window parameter w in function dtw(s,t,w) is the parameter that prevents big gradients.
Could I ask you a question? I just want to confirm with you, so that I correctly understand what is written in your file. You didn't implement the slope constraint (the one that prevents too much or too little gradient, namely the fifth condition in Sakoe & Chiba 1978), did you?
Sakoe & Chiba 1978: Dynamic programming algorithm optimization for spoken word recognition.
You need to learn what is a MATLAB function first.
What does MATLAB Compiler do? When I try to run the code, I get an error message for Line 1 (function d=dtw(a,b,w)).
Hi Steven du,
Yes, you are right. I forgot to update it when I extended it for multi-dimensional signals.
I thinks line 19 of dtw.m
Ought to be ns=size(s,1)
Hi Leonel Rozo, it now supports multidimensional signals.
Does it consider multidimensional signals?
Minor bug fixed according to the suggestion from Steven du, NTU. Thanks.
added Matlab compiler as required products
removed mex file