# ML questions: here or at Data Science?

Should questions about machine learning (ML) be asked here or at Data Science.SE? When Data Science was still a proposal, the discussion indicated these questions should be asked here. Later discussions indicated the two sites should be merged.

The tour page says "Cross Validated is a question and answer site for people interested in... machine learning." This is also a fully graduated (i.e. no longer in beta) site. However, the Data Science tour page says "Data Science Stack Exchange is a question and answer site for Data science professionals, Machine Learning specialists, and those interested in learning more about the field." A mod there claims that CV is the right site for statistics, not ML questions, but that DS is for ML questions.

As of this posting, the ML tag has 4,128 questions here and 498 there, which might suggest they fit better here.
However, the opposite conclusion might be reached by noting that under "all questions" the totals are 65,571 here and 1,505 there so ML questions are 6.3% of questions here and 33% of questions there.

Where do ML questions belong?

• It's fair to ask a mod what is on topic on their site, but you should give no weight to their characterization of another site (such as this one), especially when it is obviously at odds with the evidence. – whuber Oct 11 '15 at 14:40
• If you're referring to the dialogue in the comments here, that's not quite what the Data Science moderator said. – Scortchi - Reinstate Monica Oct 11 '15 at 15:31
• @Scortchi, it is a rather curious comment from him. He is a valued member here, as well as a mod there. Although being a mod there implies a certain level of commitment, he has more posts here (Q: 13, A: 164) than there (Q:1, A:87), & has more rep here than there. OTOH, none of his Qs here were tagged w/ ML, but his Q there was. It would be interesting for him to contribute to this thread. – gung - Reinstate Monica Oct 11 '15 at 20:13
• I agree with @Scortchi: even the apparently strongly worded referring to CV as "the SE site for ML" is simply wrong seems fine to me, if we take the emphasis to be on the "the" (which seems natural). It reads more naturally in context, as the reply to a message that says "[CV] is the SE site for ML", which was itself posted to suggest that a post was off-topic on Data Science SE (which fairly clearly it wasn't). I didn't interpret any aspect of that comment as "anti-CV" nor was it in any way opposed to ML activity on CV (if anything it was encouraged), just that ML is on-topic on DS too – Silverfish Oct 11 '15 at 20:35
• The answer to "ML questions: Here or at Data Science?" may well be both, possibly depending on context. I think this was the point that the moderator there was making. – Silverfish Oct 11 '15 at 20:39
• @Silverfish I would hope so, but I don't read it that way. – Glen_b Oct 11 '15 at 21:53
• It's even worse for ML questions about NLP, as there is a 3rd fully overlapping SE, linguistics (and some content also end up on SO and the CS SE). – Franck Dernoncourt Oct 12 '15 at 5:55
• There shouldn't be two separate sites. Problems like this will keep coming up while there are. – conjectures Oct 12 '15 at 8:24
• It turns out @AlexandrBlekh does think machine learning questions are a better fit to Data Science than to CV - see here. Fair enough; at any rate I think his comment still has to be read in the context of his responding to the perceived implication that they're off-topic on DS. – Scortchi - Reinstate Monica Oct 12 '15 at 10:08
• Haven't really been following it (casual visitor here at CV), but given that most machine learning technics are statistical in nature, distinguishing between stats and machine learning doesn't make sense to me. – James Kingsbery Oct 23 '15 at 21:21

Machine learning is on-topic here at Cross Validated—see What topics can I ask about here?, What is on topic on Cross Validated?, Are the “Machine Learning” questions on topic?, & Is machine learning a part of statistical analysis? & Should Machine Learning SE be merged with CrossValidated?. That doesn't seem to have ever been contested. (It's interesting to note that a Machine Learning SE site never took off, & had to be merged with CV— see Incoming… Machine Learning Questions.)

Though Data Science's What topics can I ask about here? isn't filled in yet, the tour page you link to & the 500 questions with the machine-learning tag (which don't seem much different in general from machine learning questions on CV) suggest machine learning's also on-topic at DS. When it comes out of beta, and as the no. users increases, I'd expect what's on-topic to be defined more clearly, & more narrowly.

So there's currently an overlap, but it's a little premature to decide what to do about it: whether it's tolerable, or whether machine learning questions should be migrated to one or the other site. Personally, I'd prefer them to stay here: at the least, there are many problems, techniques, & theories that it'd be impossible to allocate exclusively to one of the disciplines of Machine Learning & Statistics; at the most, Machine Learning & Statistics aren't two distinct disciplines but two cultures within the same discipline, between which we shouldn't be building artificial barriers.

Your "[...] Cross Validated (an ML reference) [...]" would come as a surprise to many statisticians.

• "When it comes out of beta..." -- when and if, I guess. It doesn't look very promising at the moment. – amoeba Oct 12 '15 at 12:47
• † I fortunately get to talk to a lot of machine learning experts, and a lot of people who study statistics at an advanced/graduate level (but who may not be among the best statisticians out there). Folks in the former group generally seem to take an "of course" attitude to the concept of cross validation, while those in the latter often seem to be lacking the concept altogether. Surely, some statisticians use the concept regularly in their work (it's so useful), but it doesn't seem to be as core to the field as it seems to be in ML. If it's offensive, I'll delete it. – WBT Oct 12 '15 at 15:11
• @amoeba: Well, if DS withers, that'll raise different questions. But let's hope it thrives, & is able to focus on a distinctive set of topics that aren't covered elsewhere. – Scortchi - Reinstate Monica Oct 12 '15 at 16:04
• @WBT: Seminal references are Mosteller & Tukey (1968), "Data analysis, including statistics", in Lindzey & Aronson (eds) ,Handbook of Social Psychology, 2, Stone (1974) "Cross-validatory choice and assessment of statistical predictions", JRSS B, 36, 2, & Geisser (1975) "The predictive sample reuse method with applications", JASA, 70, 350 - noting the authors & journals suggests cross-validation wasn't thought to be in any way outside the mainstream of Statistics. I'm sure no-one's thin-skinned enough to be offended; though in truth it's sometimes a little trying for ... – Scortchi - Reinstate Monica Oct 12 '15 at 16:05
• ... statisticians when it sometimes seems that anything more advanced than an t-test gets claimed by newer, more fashionable disciplines. At any rate my point's that we can't decide, & oughtn't to try to decide, who "owns" cross validation, along with many other techniques. – Scortchi - Reinstate Monica Oct 12 '15 at 16:09
• @Scortchi: You're right in that we shouldn't decide who "owns" cross validation or other techniques; we should all use what works best and seek to share & spread new ideas that work well. That I personally have for the most part only encountered the concept of CV in ML settings and not in stats settings outside ML (even encountering the notable absence of CV in such settings) may be just a fluke of my own experience. – WBT Oct 12 '15 at 19:33

It says so on StackOverflow, so it must be true!

I think that Data Science as a field is quickly sliding into obscurity in terms of applied statistics. It's becoming a hybrid of IT and business analytics. The data scientists are not able to answer AI and ML questions anymore, so they increasingly flow into CV. I see a steady stream of AI questions not only incoming, but also being properly answered. We should keep answering them so that statistical learning aspects of AI will be taken care of here.

HBR started the data science fad in the industry with the article "Data Scientist: The Sexiest Job of the 21st Century." Interestingly, they bent the quote from Google Chief Economist Hal Varian that "the sexy job in the next ten years will be statisticians." Now there was an article in January 2017 in Forbes "Can AI Make The Sexiest 21st Century Job Obsolete?." So, the industry folks went full circle on data science thing :)

• The same thing happened with "Big Data". My job title is Data Scientist : ( – Matthew Drury Oct 26 '17 at 2:43
• "the same thing" is not a bad thing, because the business problem did not disappear, it just moved under IT organization, where it is being solved. a job title can be changed. talk to your boss and HR can be asked to put a new title if necessary. – Aksakal Oct 26 '17 at 3:09
• I was just Kidding Akskal, I'm fine with being a Data Scientist. – Matthew Drury Oct 26 '17 at 3:22