The SE team apparently completely ignored us and the previous consensus and allowed Machine Learning proposal to reach private beta.

And as expected, till now all questions I managed to sneak seems to would be perfectly fit here. Moreover, it has a very poor stats and it is likely to be shut down, according to its system message:

enter image description here

So, do you think ML.SE should be merged into CV?

  • $\begingroup$ I agree. I first thought that it would be good idea to have a separate site, but now it seems that ml.SE has not attracted enough experts (but mostly enthusiasts like me) ... I shouldn't have signed up :( $\endgroup$ – steffen Feb 16 '12 at 22:27
  • 2
    $\begingroup$ The active participating experts have refused such a merge multiple times for reasons I do not understand. So I guess even if such a merge would happen, they would wander off elsewhere ... $\endgroup$ – steffen Feb 16 '12 at 22:31
  • $\begingroup$ If it's likely to be shut down, what is the point of a merger? $\endgroup$ – whuber Feb 16 '12 at 22:38
  • 2
    $\begingroup$ @whuber I assume the benefit is that we would inherit their questions and answers, which would seem to be better than letting them drift into oblivion. $\endgroup$ – deinst Feb 16 '12 at 23:06
  • 4
    $\begingroup$ I guess it depends on their quality. Beta is a time when the community is just figuring things out and a lot goes on that would be unacceptable in a more mature setting like ours. I do not relish the idea of reviewing hundreds of threads for suitability to our site. Presumably, that community has good reasons for not wishing to post their questions here, which I take as prima facie evidence that many of them might not be appropriate. $\endgroup$ – whuber Feb 16 '12 at 23:12
  • $\begingroup$ You should please post your comments as replies, @Steffen, to make them more visible as such and to enable the community to vote them up or down. Now that the question is on meta (thank you for that suggestion!), voting will have no effect on anybody's reputation: its purpose is to get a sense of the community. $\endgroup$ – whuber Feb 16 '12 at 23:14
  • $\begingroup$ Sorry all for posting on main site -- too many tabs :[ $\endgroup$ – user88 Feb 17 '12 at 0:43
  • $\begingroup$ I'd like to add that the ML site's community is very very resistant to merging into CV. $\endgroup$ – Aarthi Feb 17 '12 at 18:37
  • $\begingroup$ @Aarthi, it would be best to summarize that sentiment here directly in this thread. The private beta is blocked to the general public (although it appears by the message I received trying to access the site that won't be the case in a few days?) $\endgroup$ – Andy W Feb 17 '12 at 19:29
  • $\begingroup$ @AndyW Oh! Whoops a doo -- I fixed that link so that y'all can see the conversation now. :) $\endgroup$ – Aarthi Feb 17 '12 at 19:36
  • 4
    $\begingroup$ @Aarthi I think I detect a great overlap among the objectives of the questions asked here and at ML; the resistance seems to come from an aversion to the use of statistical thinking among some vocal members of the ML community. That's too bad; I would love to see novel methods and concepts applied to problems of data analysis and statistics. The complaints that we don't have enough ML questions (or answers) here seem rather tautological and self-fulfilling: "I don't want to contribute ML Q's or A's to the stats site because it doesn't have many such threads." Gosh. $\endgroup$ – whuber Feb 17 '12 at 20:42
  • 4
    $\begingroup$ Looking over their complaints about merging, the issue seems to be that while some of ML overlaps w/ stats, other ML issues overlap elsewhere, & they want the topic brought together in one place w/o the fragmentation that might otherwise occur. That's a reasonable point. I wonder if SE can create a system where some people could log on, not to a specific site like CV or SO, but to an account that returns a join from multiple sites. This would allow them to keep tabs on a topic that spills over w/ multiple sites. $\endgroup$ – gung - Reinstate Monica Feb 18 '12 at 23:12
  • 3
    $\begingroup$ @mbq, I stumbled upon this today, r-statistics.com/2010/07/… when stats.SE & Metaoptimize was only at the beginning.... IMHO SE somewhat lost the battle as it seems that most of the ML community is now at metaoptimize and i think that stackexchange is doing the same mistake again ... $\endgroup$ – Dov Feb 20 '12 at 20:08
  • 2
    $\begingroup$ @Dov your hyped metaoptimize has created 1473 questions since the start around july 2010, meanwhile stats.stackexchange created 7962 questions. The quota of roughly 1473 / (20 months * 30 days)=2.45 questions per day aligns pretty well with my experience from the rapidminer-forum that the ML - community is splitted up way too much. $\endgroup$ – steffen Feb 21 '12 at 7:44
  • 1
    $\begingroup$ So creating a new site stating "this is the one everyone should go to" is like setting a standard when already five standards exist. The result is not one, but six standards ;). $\endgroup$ – steffen Feb 21 '12 at 7:49

Yes, I agree that the site should be merged into stats.SE, but no, I do not think that the ML folks will be happy with that.

Frankly, when I heard about crossvalidated the first time, I hoped that it would contain more ML then it currently does. It is certainly better now, but new ML users coming to this site will still mostly see stats-related questions. A little bit disappointed by the slow growth (of ML), I signed up for ML.SE.

But now, looking at the questions and answers in the beta, I cannot get rid of the feeling that I had earlier, that most ML appliers (but certainly not all) on stackoverflow AND ML.SE are actually programmers who get excited by ML (which includes me), but cannot considered necessarily as "experts".

So ... I think that ML should grow under the corresponding tag here on crossvalidated (and not on stackoverflow) until the critical mass is reached. Then one might consider another proposal, but not now.

HOWEVER, seeing that the active ML experts on ML.SE are fighting such a merge with tooth and nail, I am afraid they will wander off when the rebel colony is forced back into the empire (Machine Learning is a duplicate of CrossValidated). All in all, I do not understand why they think that "seeding" ML questions here on crossvalidated.com is wrong but doing so on ML.SE is right.

Beside, I think that the stats- and ML-folks should learn more from each other. E.g., a little more statistical rigor in ML-validations will not hurt ...

  • $\begingroup$ "I think that the stats- and ML-folks should learn more from each other" -- that for sure! but ML should have there own site... $\endgroup$ – Dov Feb 20 '12 at 21:40
  • $\begingroup$ @Dov yes, they will. But this is like the first settlers reaching North America and before they even have built the first city stating that they want to be independent of Britain. I think the time has not come yet. $\endgroup$ – steffen Feb 21 '12 at 7:52

Could it be that the reason the the ML site is not going to well is because of meta optimize which handles these ML questions? It has a big community outside SE, as mathoverflow does as well.

I have found ML questions hard to place in SE as well. It does not even fit into scicomp.stackexchange.com/ that well. But people are right now placing questions about ML in sites like CV. I posted an ANN question a couple of days ago here because I am not part of the beta for ML, and have not had an answer.

  • 3
    $\begingroup$ Ok, I agree that this is not trivial to guess that this site is the place for ML on SE. But there are better ways to fix that than making a new site with no visibility. Moreover this does not really matter since people come to SE not by reaching list on stackexchange.com and picking site which matches their questions but by discovering a site which already have questions similar to their via Google. $\endgroup$ – user88 Feb 20 '12 at 17:43
  • 1
    $\begingroup$ @mbq, Trust me, what that Vass say is the reason!.. and please let me add something: I see that you have a high reputation here and I see that people are tend to listen to your advises...but saying (over and over) that ML and Stat.SE can leave in the same place - IMHO is sort of a miss-understanding of the ML material...I think that "newbies" at ML could be mislead by the answer at stat.SE and your statements here... I might be wrong - but PLEASE think about it. $\endgroup$ – Dov Feb 20 '12 at 21:50
  • $\begingroup$ @Dov I'm afraid I don't comprehend your comment. $\endgroup$ – user88 Feb 20 '12 at 22:18
  • 1
    $\begingroup$ @Vass As I have posted in a comment above, metaoptimize created a whopping amount of 1473 questions in 600 days meanwhile stats.SE created 7962. So metaoptimize may be far from small, but it is certainly not big. $\endgroup$ – steffen Feb 21 '12 at 7:59
  • 1
    $\begingroup$ @steffen. but in metaoptimize they are all ML question where Stat.SE has only 430 on ML. stats.stackexchange.com/tags/machine-learning/info while about 20% unanswered stats.stackexchange.com/questions/tagged/… $\endgroup$ – Dov Feb 21 '12 at 13:26
  • 3
    $\begingroup$ @mbq although Dov's English isn't that good, I think the point he's trying to make is that ML and stats have separate identities. I have to say I agree - I struggle with the content on here but am very comfortable with questions on MO, and was also on the ML.SE beta. $\endgroup$ – tdc Feb 21 '12 at 13:27
  • 1
    $\begingroup$ @steffen agreed on that point - but I think this was a opportunity for ML.SE to become the place for ML questions, rather than being indicative of a general lack of interest in the area. I think once people have "bought in" to the SE idea, they're likely to have accounts in many places. However having an account on a standalone clone clearly doesn't have the same appeal. $\endgroup$ – tdc Feb 21 '12 at 13:29

I don't agree with the merge, and my reasoning is this: there are quite a few places on SE where one could ask ML-related questions:

  • programmers.stackexchange.com
  • stats.stackexchange.com
  • quant.stackexchange.com
  • math.stackexchange.com
  • cstheory.stackexchange.com

I personally immediately got confused as to where should I ask my ML questions for them to be best answered. And I was extremely relieved when I've heard about a dedicated MachineLearning site.

Although, unfortunately it doesn't go as well as some people expected it to! :)

  • 2
    $\begingroup$ Again, this not a site on which you just "can" ask about ML, this is a site which is explicitly devoted to ML (among stats and visualization). This is like arguing that there should be C++.SE because you can ask C++-featuring questions on circa half of SE sites. $\endgroup$ – user88 Feb 20 '12 at 10:32
  • $\begingroup$ Well, maybe. But to me it's more like Game Development being a separate one, and Security being a separate one - why not just ask everything on stackoverflow, eh? :) But I do understand your point and still don't agree - perhaps because I'm not used to SE. $\endgroup$ – andreister Feb 20 '12 at 11:10
  • $\begingroup$ Because questions about game AI, best looking stone textures, password strength or deploying fake GSM station are not about programming. $\endgroup$ – user88 Feb 20 '12 at 12:51
  • 1
    $\begingroup$ Similarly as questions about machine learning are not necessarily for "statisticians, data analysts, data miners and data visualization experts" per se, no? $\endgroup$ – andreister Feb 20 '12 at 13:27
  • 1
    $\begingroup$ The official scope is in the FAQ. This tagline is very hard to change, thus its far from perfect. $\endgroup$ – user88 Feb 20 '12 at 14:49
  • $\begingroup$ How would you propose modifying the FAQ tagline, @mbq? (As you know, mods can easily edit it, so I presume by "hard" you mean that significant changes are usually discussed first.) $\endgroup$ – whuber Feb 20 '12 at 19:14
  • 1
    $\begingroup$ @whuber I meant the description from SE.com and about page. Those are controlled by the team only; an extensive discussion is of course also an inevitable step. $\endgroup$ – user88 Feb 20 '12 at 19:48
  • $\begingroup$ @mbq OK, got it. But my question still stands: you have said the tagline is "far from perfect." Would you like to initiate a conversation (chat or meta) about how to improve it? Or, imperfect as it may be, perhaps it's passably ok? :-) $\endgroup$ – whuber Feb 20 '12 at 19:51

You must log in to answer this question.