0
$\begingroup$

I would like to visualize a large amount of events composed of time serie windows. A typical event would be:

typical event

Problem is, my events are not synchronized, and so if I plot them all, it would look like:

enter image description here

Question
Is there any way to visualize all my events so I can see their original/"typical" shape (preferably in the time domain) despite their unsynchronization ?

What I have tried so far:

  • Visualize features: approach is good but I have to guess what I am looking for.
  • Synchronized events: in the example above it might be possible, but for some other cases it won't be. (such as multiple peaks spaced with different time lapses)
  • Use statistical values to visualize: that was my first shot, to plot not all events but only their mean, median, p95,... Problem is, using this approach gives a deformed rendering of the data because of the unsynchronization:

enter image description here

What I have thought about

The main problem of applying the mean, p95 or whatever statistical function is that it is applied along an axis that is not synchronized between events. I don't know if it exists or if it is even feasible, but I was thinking about an approach that would display those statistical values checking their temporal neighbors ? I know Dynamic Time Warping deals with this unsynchronization but I am not sure if and how I could use it to plot my events in the way I'd like.
An idea of a possible output could be (any other output idea is welcomed!):

enter image description here

Any help, idea, would be greatly appreciated ! Thanks

$\endgroup$
4
  • 1
    $\begingroup$ DTW is a good solution for this problem. This post might help you ealizadeh.com/blog/introduction-to-dynamic-time-warping $\endgroup$
    – WBM
    Feb 26, 2021 at 10:39
  • $\begingroup$ Thanks @WBM, how would you do use DTW to display a graph like the last one ? Have you seen raw data visualization based on DTW ? $\endgroup$
    – etiennedm
    Feb 26, 2021 at 10:43
  • $\begingroup$ So you use DTW to cluster the clips in time (or warp them so they're aligned) and then you create an average of those clusters which would be your ideal_sig in the final image $\endgroup$
    – WBM
    Feb 26, 2021 at 10:46
  • 1
    $\begingroup$ Thanks, I think I see your point, I will dig into it and try to build a function that does that $\endgroup$
    – etiennedm
    Feb 26, 2021 at 11:20

1 Answer 1

0
$\begingroup$

As far as estimating similarity between the time series series there is a variety of methods you may want to investigate. Some of those are:

  • cross correlation: this will be affected by the amplitude and will not be able to estimate lagged correlations, prone to noise.
  • coherence: normalised frequency based correlation (cross-spectrum), not prone to amplitude or noise.
  • wavelet coherence: similar to above but based on wavelet transformations instead of STFFT.
  • dynamic time warping: measuring similarity between two temporal sequences, which may vary in speed.
$\endgroup$
4
  • $\begingroup$ Thanks @hH1sG0n3, I am looking for a tool that precedes the investigation of similarities, a tool that could point me where to look for similarities into a large amount of data by visualizing them in a specific way. I don't know if such a tool/plot exists or if as you proposed, I have to directly dig into all possible features that extract similarities. $\endgroup$
    – etiennedm
    Feb 26, 2021 at 11:24
  • $\begingroup$ Coherency metrics can be calculated across multiple events. I thought this may be useful since this can be considered an event related analysis? $\endgroup$
    – hH1sG0n3
    Feb 26, 2021 at 12:47
  • $\begingroup$ Yes, definitely it is related, thanks. But I am looking for a way to visualize all my events so I can see their original shape (preferably in the time domain) despite their unsynchronization. The example I have used is simplified and in the real case I have no clue about the shape. I think I know how to use your metrics and as they are, I don't think any directly achieve what I am looking for, do they ? Sorry if it wasn't clear in the question. $\endgroup$
    – etiennedm
    Feb 27, 2021 at 9:09
  • $\begingroup$ I have updated the question, I hope it is more clear now $\endgroup$
    – etiennedm
    Feb 27, 2021 at 9:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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