stumpy icon indicating copy to clipboard operation
stumpy copied to clipboard

Add Contrast Profile Tutorial

Open seanlaw opened this issue 4 years ago • 7 comments

This paper is here

The supporting site is here

The main goal of a contrast profile is to annotate each subsequence of a time series so that you can discover if there are "good" subsequences that can be used for time series classification (see Definition 6 in the paper above). In order to extract these subsequences, you need to have 2 time series. One time series, , and another time series, . It is recommended that you compute the the Contrast Profile only when you believe that the two following assumptions are likely to be true:

  • contains at least two behaviors that are unique to the phenomena of interest.
  • contains zero behaviors of interest.

seanlaw avatar Sep 08 '21 14:09 seanlaw

The following notebook reproduces the PDF which is contained in "ContrastProfile_GettingStarted.zip" contrast_profile_notebook.pdf

JaKasb avatar Jun 02 '22 16:06 JaKasb

Thanks @JaKasb! This is a great start. Would you care to contribute a full tutorial for this?

seanlaw avatar Jun 02 '22 19:06 seanlaw

After re-reading the paper, I have come to the conclusion that I can't imagine a specific topic that needs an extra tutorial. The paper already explains online-updates, k-Platos, and comparisons of Contrast-Profile vs Shapelets/HMM/LSTM.

JaKasb avatar Jun 03 '22 16:06 JaKasb

In case it matters, our tutorials generally reproduce one or more of the figures in the paper and tries to elaborate on any insights/thoughts beyond what was provided in the paper that might not immediately be obvious to non-experts. Sometimes, reading the same concepts but explained in different ways may help others digest the information. That's usually the goal of our STUMPY tutorials. Code is one component but educating users on how to think about and interpret their results is equally as important. It may help provide more perspective.

seanlaw avatar Jun 03 '22 19:06 seanlaw

Is there still demand to reproduce this paper? Basic idea looks not complicated.

ken-maeda avatar Mar 01 '23 08:03 ken-maeda

Is there still demand to reproduce this paper? Basic idea looks not complicated.

Sure! If you'd like to work on a docs/Tutorial_Constrast_Profile.ipynb then I can review it. I think the code should be straightforward but providing a clear explanation is more important

seanlaw avatar Mar 01 '23 16:03 seanlaw

@seanlaw I uploaded it. When you have time, please give me feedback.

ken-maeda avatar Mar 03 '23 08:03 ken-maeda