10
$\begingroup$

I cannot see any reason for the tag to exist on CV. It would seem this is very specifically about code, and not directly statistical/ML code.

But I've never used pandas.

EDIT:

One point I did not make clear enough in the original post is that, as far as I know, pandas is not a package containing any implementation of statistical methodology (or machine learning); it's just a commonly used tool for storing and organizing data. It is my understanding that a question about, say, scikit-learn could easily be on topic, especially if the question focused on the statistical methodology behind the tools in the package. But it seems like a question about SQL is almost surely off topic. For me, a tag like is much closer to than it is to .

But maybe pandas does more than I recognize?

| |
$\endgroup$
  • 3
    $\begingroup$ Would the same argument apply to scikit-learn tag? That one has over 500 questions by now and is very active. $\endgroup$ – amoeba Mar 6 '17 at 16:02
  • 3
    $\begingroup$ Meanwhile, I wrote a wiki excerpt for pandas. $\endgroup$ – amoeba Mar 6 '17 at 16:02
  • 3
    $\begingroup$ Speaking about Python tags, there is also matplotlib. But if we start deprecating tags like that, a lot of R tags will also need to be deprecated (e.g. ggplot2). $\endgroup$ – amoeba Mar 6 '17 at 16:04
  • 3
    $\begingroup$ Given there is [R] tag (which is the top tag on the site), I don't see why [pandas] is fundamentally different. The same caveats expressed in [R] tag excerpt should apply to other tools. $\endgroup$ – user56674 Mar 6 '17 at 16:05
  • $\begingroup$ @NormalHuman [r] corresponds to [python], not to [pandas]. $\endgroup$ – amoeba Mar 6 '17 at 16:06
  • 5
    $\begingroup$ I agree w/ @amoeba. Despite our general software policy, I do see a place for tags like r, python, etc. OTOH, sub-software tags are ambiguous. It seems clear to me that lme4-lme has sufficient statistical content, but I'm not sure if, say, ggplot2 does, & I'm not python savvy enough to know if pandas is more like ggplot2 or lme4-lme. I suspect scikit-learn is like lme4-lme. (On the 3rd hand, if we do start depreciating these, it's going to be a lot of work, & we don't seem to make a lot of progress on tag management tasks.) $\endgroup$ – gung - Reinstate Monica Mar 6 '17 at 17:34
  • 1
    $\begingroup$ @gung I think [lme4-nlme] is an important tag because these packages do not just implement well-defined algorithms, they almost define algorithms, at least to a certain extent; Pinheiro & Bates have a whole book essentially on lme4, AFAIK. In contrast, I don't see much statistical content in [scikit-learn], because it is a collection of well-established statistical algorithms; the developers of the package are not the authors of these algorithms. And I fully agree that there is no statistical content in [ggplot2] or [matplotlib]. As well as in [pandas]. $\endgroup$ – amoeba Mar 6 '17 at 23:00
  • $\begingroup$ @gung [continued] In fact, deprecating these tags can be really easy if we simply declare them synonyms of the master language tags. I.e. pandas, scikit-learn, and matplotlib can become synonyms of [python]. And ggplot2 can become a synonym of [r]. This might be a reasonable approach, perhaps worthy of a separate Meta discussion. I don't have a strong opinion either way. $\endgroup$ – amoeba Mar 6 '17 at 23:02
  • 3
    $\begingroup$ In reply to your Edit: this makes sense. Would you then suggest to eliminate the [sql] tag? What about [latex] tag that was discussed another day on Meta? $\endgroup$ – amoeba Mar 7 '17 at 11:58
  • $\begingroup$ @amoeba, you could post an answer proposing to make [pandas] a synonym of [python] so that it could be discussed & voted on explicitly. $\endgroup$ – gung - Reinstate Monica Mar 7 '17 at 12:21
  • $\begingroup$ @gung I am not sure where I stand on this issue. If scikit-learn can be left alone, then I am not sure that pandas can't. $\endgroup$ – amoeba Mar 7 '17 at 14:05
  • 2
    $\begingroup$ The edits seem to suggest that data management and manipulation have nothing to do with statistics and machine learning. In my experience, 80% to 90% of the effort on many statistical projects consists of data processing and the details of data management can have important statistical ramifications. Consequently I have little difficulty imagining how even certain SQL-related questions (obviously not all of them!) could be considered relevant and on topic. $\endgroup$ – whuber♦ Mar 8 '17 at 17:04
  • 1
    $\begingroup$ @whuber: I agree that data manipulation and some aspects of data management are very appropriate questions for CV. But if we allow questions about software for data management, I think that's an extremely slippery slope. To illustrate my view, I think the question "what is a subset of my complete data that I can use to answer my question of interest?" is about data manipulation/management that's definitely on topic, but "how do I use filters in sql?" is also about data manipulation but not on topic. I see the first question as a question about sampling, the second a question about SQL. $\endgroup$ – Cliff AB Mar 8 '17 at 21:56
  • 1
    $\begingroup$ I agree w/ @CliffAB here. I am hard pressed to imagine any question that is actually about SQL that would be sufficiently about statistics / machine learning, etc., to be on topic here IMO. I do recognize that 80-90% of my effort on statistical projects consists of things other than pure stats. Eg, I spend a certain amount of time fussing w/ Outlook (which I am required to use) to manage email threads among multiple collaborators. This is a continuing source of frustration for me & accounts for a real (non-0) amount of my time, but I don't think such questions should be on topic here. $\endgroup$ – gung - Reinstate Monica Mar 9 '17 at 20:30
  • 1
    $\begingroup$ @amoeba: ggplot2 is about statistical visualization, which is certainly on-topic, when not only about coding. The exactly same seems to apply to tags like lme4-lme $\endgroup$ – kjetil b halvorsen♦ Mar 10 '17 at 10:08
7
$\begingroup$

Tags serve several purposes. Software tags inform us that

  • the original proposer is interested in explanations or solutions that are, or can readily be expressed in, certain software platforms;

  • any code offered within the question or answer should be pretty-printed according to the specified syntax.

For instance, in any thread with an tag, the default formatting for code blocks will assume it's R. (I don't know what happens when multiple software platforms are tagged.)

However, because this is a site about methods and concepts rather than software per se, we welcome answers that are independent of any specific platform and we accept answers that might be expressed in a language other than specified in the tags. We do require that questions and answers can be understood by people who are not conversant with any particular programming language. In particular, a question or an answer that consists entirely of code--even when it's in a tagged language--almost surely will be deleted.

| |
$\endgroup$
  • $\begingroup$ This all sounds very reasonable, but I am not sure if you argue for or against [pandas] tag. Or e.g. [matplotlib] or [ggplot2] (to name some examples from the comments above). $\endgroup$ – amoeba Mar 6 '17 at 22:55
  • $\begingroup$ @amoeba I had assumed, perhaps incorrectly, that the conclusion to be drawn from a list of positive attributes, accompanied by no negative attributes, would be clear. $\endgroup$ – whuber♦ Mar 6 '17 at 23:04
  • 3
    $\begingroup$ Hmmm, I actually inferred the opposite then. I was thinking that these are potential positives of software tags such as [r], [python], etc, that do not seem to apply well to [pandas] & thus [pandas] should not be afforded the same protection. $\endgroup$ – gung - Reinstate Monica Mar 6 '17 at 23:07
  • 1
    $\begingroup$ @gung This was my interpretation too. $\endgroup$ – amoeba Mar 6 '17 at 23:13
  • $\begingroup$ @Amoeba Could you explain what it is about Pandas that would distinguish it from, say, R as a tag? $\endgroup$ – whuber♦ Mar 6 '17 at 23:30
  • 3
    $\begingroup$ See my two last comments (addressed to @gung) under Cliff's question. I'd say that [pandas] (i) does not have any statistical meaning, (ii) might falsely suggest to somebody that questions about coding something using pandas are on-topic, (iii) does not help syntax highlighting because one can simply use [python] instead (and pandas tag is probably not associated with python syntax anyway, even though this can be fixed). Therefore one could suggest to synonymize pandas with python, and similarly for most other package tags with rare exceptions for particularly prominent/important packages. $\endgroup$ – amoeba Mar 6 '17 at 23:36
1
$\begingroup$

Consider some existing tags for (1233) libraries:

There are several possible things we can do with them:

  1. Do nothing: leave all these tags alone.

  2. Make all these tags synonyms of . This is pretty radical, some of these tags are widely used.

  3. Keep the tags about statistical libraries, but get rid of the tags about non-statistical libraries; that is what you are suggesting in your edit. However, I find it difficult to draw a line here. [scikit-learn] is clearly statistical. You say that [pandas] is not, but it is about data storage and manipulation which can be very pertinent to statistical analysis and on-topic; also, it includes some statistical processing routines. [matplotlib] appears entirely non-statistical, however it is about data visualization which is on-topic too.

Given these considerations, I am inclined to prefer approach #1.

One thing we could do to address your concern about off-topic questions, is to add a cautionary note into tag excerpts warning people that programming questions are off-topic. E.g. [pandas] wiki excerpt currently reads

Python library for data manipulation, implementing R-style data frames.

but we could add something like

Python library for data manipulation, implementing R-style data frames. Programming questions about Pandas are off-topic unless they have statistical content.

| |
$\endgroup$
  • 1
    $\begingroup$ I (not surprisingly) vote for (3). $\endgroup$ – Cliff AB Mar 9 '17 at 13:21
  • $\begingroup$ I would be inclined to prefer option #3. Certainly data storage and manipulation can be very pertinent to statistical analysis, but a whole lot of other things are pertinent that are off topic. As our help states, "if ... [your question is] about ... routine data processing [it is off topic]". $\endgroup$ – gung - Reinstate Monica Mar 9 '17 at 13:24
  • $\begingroup$ If there is a close mapping between the library and a particular area of statistics would it not be better to make it a synonym of the statistical concept? So for example ggplot2 would map to data-visualisation. Sorry I cannot give examples for the python ones as I do not use python $\endgroup$ – mdewey Mar 9 '17 at 13:34
  • $\begingroup$ @CliffAB Can you name all tags from my list that you would get rid of? Or is it only [pandas]? Let's try to see if we can work our some boundary that is self-consistent. $\endgroup$ – amoeba Mar 9 '17 at 13:52
  • $\begingroup$ @gung Can you name all tags from my list that you would get rid of? Or is it only [pandas]? Let's try to see if we can work our some boundary that is self-consistent. (Also, maybe we could make/discuss a similar list of R package tags to put this into a broader context.) $\endgroup$ – amoeba Mar 9 '17 at 13:52
  • $\begingroup$ @mdewey This did cross my mind too. Using this logic, e.g. [theano] and [tensorflow] should be mapped to [neural-networks]. However, it's not clear to me where to stop. Would you map [lme4-nlme] with its 1.5k threads to [mixed-models] (2.2k)? I would say such a mapping would be enormously detrimental for the site. $\endgroup$ – amoeba Mar 9 '17 at 13:56
  • $\begingroup$ @amoeba: having never used python, I feel like my vote about which tags should go and which should state should have exactly 0 weight. $\endgroup$ – Cliff AB Mar 9 '17 at 14:03
  • $\begingroup$ @CliffAB That's fair enough. You are using R, aren't you. How about you name some existing R package tags that you consider non-statistical enough to get rid of? ggplot2? dplyr? sweave? caret? $\endgroup$ – amoeba Mar 9 '17 at 14:12
  • 1
    $\begingroup$ As far as [lme4-nlme] is concerned I agree that you cannot now put the toothpaste back in the tube. $\endgroup$ – mdewey Mar 9 '17 at 14:25
  • $\begingroup$ I prefer #1, and suggest the 'off-topic' to be uppercase: 'OFF-TOPIC'. Sometimes, screaming helps :). $\endgroup$ – Andre Silva Mar 9 '17 at 14:35
  • $\begingroup$ For the case of R, [ggplot2] questions could be on topic as data visualization, [lme4-lme] & [caret] can certainly have sufficient statistical / machine learning content; OTOH, I am hard pressed to imagine a question that is actually about [dplyr], [sweave] (or [latex], [sql], etc) that has sufficient stats / ML content to be on topic. I'm not sure I know Python well enough, but I suspect [scikit-learn] should stay & [pandas] should go. $\endgroup$ – gung - Reinstate Monica Mar 9 '17 at 20:19
  • $\begingroup$ @gung I can see your reasoning. I would encourage you to post an answer here elaborating on this position, so that we could discuss it & vote on it. $\endgroup$ – amoeba Mar 10 '17 at 14:56
  • $\begingroup$ @amoeba I agree with #3 but I'd keep [tensorflow]. It becomes pretty popular and it is something more then "just" a library (something like lme4). $\endgroup$ – Tim♦ Aug 9 '17 at 14:02

You must log in to answer this question.

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