Dynamic Time Warping

[wangkev]

Thank you for the awesome library!

Wanted to follow up the discussion of usage of matrix profiles with dynamic time warping for motif discovery. Seems like the there was recently a paper published on the subject. Is this on the roadmap to support?

Past Discussions:

• Stumpi with custom metric #233

[seanlaw]

Thank you for your question @wangkev. I am aware of this paper but this it not currently on our roadmap. Unfortunately, I have limited experience with DTW and my feeling is that, without assistance, implementing DTW is beyond the scope of this package. It seems that there are a lot of ways to optimize DTW and my fear is that we can spend all of our bandwidth on chasing a moving target.

Of course, this is my own limitation and we would welcome others to contribute to this package.

[seanlaw]

In fact, I think that this would be more appropriate as a separate package that might depend on STUMPY. While the capability might be useful/valuable to users, it is also important (from a long term maintainability standpoint) that STUMPY not try to be everything to everyone.

[seanlaw]

@wangkev If it's okay with you, I'd like to close this for now. Feel free to re-open, comment, or open up a new issue if you have any further questions.

[wangkev]

@seanlaw Understood, thanks!

[Djangu-algo]

This may or may not be helpful but at least interesting on the topic of this discussion and possible future implementation "Everything you know about Dynamic Time Warping is Wrong"

[jonahpearl]

Also interested in this...@wangkev did you ever find or make anything in python for implementing the algs in the paper?

[saeed349]

Would also love to see this implemented in stumpy