4
$\begingroup$

[The initial pun turned into a look at observable variables showing trends and clustering]

Outlier detection is a radical task in signal processing. The user Matt L. has about three times more total reputation and year reputation than the second best.

  • What are his secrets?
  • How can they benefit to the average crowd?

Nota: I was thinking about proposing a tag. There was a DSP humor cartoon contest at ICASSP 2000. This meta post is a way to test the sensitivity of the audience:

Dirty DSP talk

$\endgroup$
  • 1
    $\begingroup$ :-) Darn fine question! @Matt L. is a great resource for SP.SE users. $\endgroup$ – Peter K. Oct 20 '16 at 11:49
  • 7
    $\begingroup$ Hey, this is about me! Only saw it just now. I guess the answer is not hanging around Meta too much :) $\endgroup$ – Matt L. Oct 25 '16 at 21:16
  • $\begingroup$ A comment which deserves becoming an answer, I guess :) $\endgroup$ – Laurent Duval Oct 26 '16 at 3:43
  • $\begingroup$ Is he a John Skeet or a Don Bradman? $\endgroup$ – TRiG Nov 2 '16 at 17:20
  • $\begingroup$ No, I cannot think to that level $\endgroup$ – Laurent Duval Nov 2 '16 at 17:32
3
$\begingroup$

Answer a lot and answer well? Some statistics:

Top 20
Figure 1. Number of answers and reputation of all-time top 20 users, with a trendline fit to the data.

$\endgroup$
  • $\begingroup$ Quite impressive results. What I like with regression and intercept, is that one could get reputation without answering :) $\endgroup$ – Laurent Duval Oct 25 '16 at 13:11
  • 2
    $\begingroup$ Yes, but you may have to be in the top 20 for that! Was it 100 rep that you get if "we trust you on other stackexchange sites". $\endgroup$ – Olli Niemitalo Oct 25 '16 at 13:19
  • 2
    $\begingroup$ Nicely done, @OlliNiemitalo! Thanks for your graphs, and stirring @Laurent into action. :-) $\endgroup$ – Peter K. Oct 31 '16 at 12:21
  • 2
    $\begingroup$ Who is taking care now of @Matt L. advice: "the answer is not hanging around Meta too much" :) $\endgroup$ – Laurent Duval Oct 31 '16 at 12:25
  • $\begingroup$ Based on my observations, this would be the correct answer. Matt L answers a lot of questions, but he also provides excellent answers. I gave up trying to keep up with his reputation count a long time ago. :) $\endgroup$ – Jason R Nov 2 '16 at 13:30
  • $\begingroup$ @JasonR Let me acknowledge your authoritative comment by validating this answer. $\endgroup$ – Laurent Duval Nov 26 '16 at 9:53
3
$\begingroup$

As I am studying dimension reduction methods, @OlliNiemitalo answer triggered my curiosity. I pushed the study to rank 30, and the answer/reputation graph looks like:

Answers-reputation graph

No drastic change in the trend. However, more points cluster at the bottom. So I tried with other easy to reach variables: Reputation, #Month (of presence), #Answers, #Questions, #PeopleReached. The question/reputation graph possibly rules out the number of questions as a fine variable:

Questions-reputation graph

Then, going to 3D, we can try answer/month/reputation:

Answers-month-reputation graph

or answer/reach/reputation:

Answers-reach-reputation graph

Those 3D graphs seem to display a bended shape, which clusters when projected in 2D. The crowd of visionaries (early adopters) is quite apparent from the figures. Time matters a bit apparently. But there should be hidden variables. I hope I am not wasting your time (and mine) on this. I am not sure this deserves higher-order nonlinear dimension reduction techniques.

$\endgroup$
  • 1
    $\begingroup$ Beautiful graphs ! $\endgroup$ – Gilles Oct 28 '16 at 20:02
  • 1
    $\begingroup$ Very cool! Thanks for the analysis. :-) $\endgroup$ – Peter K. Oct 31 '16 at 12:21

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .