Dynamic Time Warping (DTW) is a way to search a time-series for a given pattern. Basically, instead of Euclidian Distance (ED) for comparing the pattern to the time-series, DTW can be used and allow for more useful matches. Typically, DTW is significantly slower than ED, if implemented naively. Luc…

I don’t understand the choice of using GitLab. This means that you won’t be able to register it in the general registry and we will never be able to do ]add UcrDtw. What is the rationale for this?

My company is using GitLab, so I have to use it.

[image] Datseris: I don’t understand the choice of using GitLab. This means that you won’t be able to register it in the general registry and we will never be able to do ]add UcrDtw . I don’t think this is true. The registry contains a git URL, but I don’t think it cares where that URL is hos…

This is correct, and things will eventually change so that Gitlab and other git platforms are supported for automatic merging into General.

[image] Datseris: I don’t understand the choice of using GitLab. This means that you won’t be able to register it in the general registry and we will never be able to do ]add UcrDtw . What is the rationale for this? I think everyone would agree that ultimately we are going to need a way to re…

I’m super glad to see more DTW support. DTW is really useful in a lot of audio processing contexts, and the other implementations don’t seem like they’re maintained, so this is a welcome development. Thanks for putting the work in!

Very interesting! I have several questions: How do you handle noisy edges, when S is a squeezed version of Q? According to the project description, Return the maximum distance encountered during the search. This allows for mapping the distance values onto a similarity range [0, 1]. What…

[image] sairus7: How do you handle noisy edges, when S is a squeezed version of Q? That isn’t covered yet, but I did find a promising paper for this use-case, which I document in an issue . [image] sairus7: How about transforming DTW distance to the correlation coefficient? The goal o…

[image] Seanny123: I’ve never heard of shapeDTW! It seems quite promising. I’ll curious how it compares to Matrix Profile, since they both seem concerned with sub-sequences. However, it does seem easier to implement than Matrix Profile… As far as I know, the main idea of this is that you can …

[image] Seanny123: First, I want to finish the multi-variate code So currently multivariate data isn’t supported? [image] sairus7: As far as I know, the main idea of this is that you can transform your series into some feature series/sequence and then optimize distance in feature space, …

[ANN] UcrDtw.jl: a "fast" implementation of Dynamic Time Warping

Community

Topic		Replies	Views
[ANN] DynamicAxisWarping.jl Package Announcements package , time-series , dsp	5	1424	August 26, 2020
Dynamic Time Warping in Julia Signal and Image Processing	7	3469	April 24, 2020
Technical question about DelayEmbeddings.jl Specific Domains	12	734	August 14, 2021
[ANN] Durbyn.jl — Time Series Forecasting in Julia Package Announcements package	66	4639	February 25, 2026
Time-series in Julia (working list) Statistics time-series	41	14528	August 14, 2025

[ANN] UcrDtw.jl: a "fast" implementation of Dynamic Time Warping

Related topics