1
$\begingroup$

I am planning to cluster a great amount of time series of different lengths into groups without using the method Dynamic Time Warping, but something else which gives a little better execution time and results. What's your opinion?


I mean that I need an algorithm, exept from DTW but a litter better than it, that measures similarity between temporal sequences -times series of different lengths- and then performs clustering using distance measures techniques as criterion. Then I will be able to perform forecasting to time series according to the train sets time series which I have clustered

$\endgroup$

closed as unclear what you're asking by Alexis, Michael Chernick, kjetil b halvorsen, mdewey, Xi'an May 10 '18 at 8:10

Please clarify your specific problem or add additional details to highlight exactly what you need. As it's currently written, it’s hard to tell exactly what you're asking. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.

  • 1
    $\begingroup$ Computing the DTW requires O ( N 2 ) in general. Fast techniques for computing DTW include PrunedDTW,[1] SparseDTW,[2] FastDTW,[3] and the MultiscaleDTW.[4][5] A common task, retrieval of similar time series, can be accelerated by using lower bounds such as LB_Keogh[6] or LB_Improved.[7] In a survey, Wang et al. reported slightly better results with the LB_Improved lower bound than the LB_Keogh bound, and found that other techniques were inefficient.[8] from en.wikipedia.org/wiki/Dynamic_time_warping $\endgroup$ – Nikolas Rieble May 7 '18 at 14:11
  • $\begingroup$ Other distance measures to use would be Euclidean Distance (ED) or Longest common subsequence (LCSS). $\endgroup$ – Nikolas Rieble May 7 '18 at 14:13
  • $\begingroup$ Please clarify your question. What do you mean by "clustering", what is your expectation on "better execution time" and for the result? $\endgroup$ – cherub May 7 '18 at 14:22
  • $\begingroup$ I mean that I need an algorithm, exept from DTW but a litter better than it, that measures similarity between temporal sequences -times series of different lengths- and then performs clustering using distance measures techniques as criterion. Then I will be able to perform forecasting to time series according to the train sets time series which I have clustered. $\endgroup$ – Nikos May 7 '18 at 14:44
3
$\begingroup$

Better results that DTW? For the problems of time series classification, there is nothing significantly better than DTW, see https://arxiv.org/abs/1602.01711

As for "better execution time", DTW with a warping constraint and LB_keogh is fast enough for virtually any task.

See http://www.cs.unm.edu/~mueen/DTW.pdf

$\endgroup$
1
$\begingroup$

You can always approximate your data if scalability is an issue.

Also similarity search indexes can speed up many algorithms.

$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.