7
$\begingroup$

I think this question was closed incorrectly: Is a strong background in maths a total requisite for ML? This asked, "Is a strong background in maths a total requisite for ML? Should I try and fill in the blanks of my maths before continuing with ML? Can self learning really work for just a developer without any hard computer science background?"

It was closed as an exact duplicate of Book for reading before Elements of Statistical Learning? This asked, "I want to digest Elements of Statistical Learning... I don't have enough knowledge to understand it. Can you recommend a book that is a better introduction to the topics in the book?"

In other words, the new question asks, "Is X something I need to do first?" The old question says "I have decided I need to use method Y to do Z. Please help me find the resources to do method Y." They seem quite different to me even if you assume X=Z and that there is only one method Y for accomplishing this. Neither my answer nor the comment by John Moeller made those assumptions, and these would not have fit the previous question. The answers to the previous question might be helpful, but they are not answers to the new question.

So, what was gained by closing this question? What was lost was the ability for people to add other answers to the new question which don't fit the old question. Please reopen this question.

$\endgroup$
5
$\begingroup$

I have voted to close the question as duplicate, because it seems to me that both say:

I have a hard time to understand ML. Can you point me to a gentle introduction (where one OP names the gaps and one does not).

We already have a couple of introductory reference questions on our site and I am a little bit tired of seeing new ones popping up again and again ... meanwhile many users seem to be eager to answer those questions instead of closing them as duplicate. Everyone has a different background, but does it really make sense to allow one question per profession / degree of knowlege ?

BUT: If the focus on the question is explicitly "Is a strong background in maths a total requisite for ML?" (so this is also the title of the question) and we downvote answers which only provide references to introductory material to Machine Learning (because this answers belongs to the current duplicate) and upvote those who provide insights to the question and/or introductory material to "learn math the right way"-material, I am totally fine reopening it.

I admit that I was too trigger happy (when using the close-cannon). Thanks for raising this !

$\endgroup$
5
$\begingroup$

Context

The duplicate question states, "I don't have enough knowledge to understand [Elements of Statistical Learning]" and asks (therefore) for more suitable introductions.

The closed question does not have such a clear focus. Its title, "How to get an introductory understanding of Machine Learning," appears to duplicate the former one. Its text asks something different, phrased as three questions. They all are variants of "Is a strong background in maths a total requisite for ML?"

I interpreted these two as broadly asking the same thing: references to resources for learning about ML for people without the requisite background.

About Duplicates

"Duplicate" questions are not supposed to be exact duplicates of each other, but only duplicates of the content: "This question covers exactly the same content as earlier questions on this topic." As far as I can tell, the two questions we are discussing are duplicates in this sense.

Closing a question as a duplicate is good for everybody:

  • It keeps topics clearly organized.

  • It enhances searching.

  • It speeds up finding answers by limiting the places one has to look.

  • It consolidates closely related material.

  • The person asking the question immediately gets answers--usually with votes and comments to help evaluate their quality or usefulness.

One motivation to keep a question open might be to help its respondents continue to accumulate reputation for their replies. This can just as well be accomplished by merging the threads. Merging, however, is more difficult to undo than closing (I am not aware of any way to completely reverse a merge) and can be performed only by moderators. Therefore it seems wise to delay merges in any case where two questions are not clearly exactly the same.

Resolutions

If this question is re-opened, I recommend it should be done so only if accompanied by edits to give it a clear focus. Depending on those edits, it perhaps also ought to be made CW, because this is one of those peripheral, possibly useful, topics that are not likely to have any objectively correct or best answer.

$\endgroup$
  • 1
    $\begingroup$ The questions are clearly similar and clearly do not cover the exact same content. That one is explicitly asking for book recommendations and the other does not mention books seems to be prima facie evidence that they are not exact duplicates. $\endgroup$ – Michael McGowan Oct 23 '12 at 14:56
  • $\begingroup$ Yes, @Michael, that's one of my main points: the concept of "duplicate" on this site does not involve exactness. One question can duplicate another--and this often happens--when they ask substantially the same question but in different contexts or with different specific applications. We don't really want to see a thousand variations of "how do I compute a Z-value for this statistic," after all! Thus, when the answers to an existing question can reasonably be construed as applying, without any special extra reasoning or knowledge, to a new question, then the new question usually is closed. $\endgroup$ – whuber♦ Oct 23 '12 at 14:58
  • $\begingroup$ "I interpreted these two as broadly asking the same thing: references to resources for learning about ML for people without the requisite background." Except that the "duplicate" is explicitly asking for books. Maybe both questions are duplicates of a hypothetical, generalized question, but that question has not been asked. If someone writes that question then maybe both of these are duplicates of the abstract question. Until then, they are certainly not exact duplicates of one another. $\endgroup$ – Michael McGowan Oct 23 '12 at 15:03
  • $\begingroup$ @Michael I have already agreed they are not exact duplicates. Where we might differ is that I perceive a request for book references as being of little or no practical consequence in differentiating questions. I view "what are some references to learn X" and "Could you recommend a book to learn X" as being substantially the same. We might ask whether it is worthwhile maintaining pairs of such questions on our site. I say not, because a future seeker of information for references to X would then have to consult two threads instead of one. $\endgroup$ – whuber♦ Oct 23 '12 at 15:11
  • $\begingroup$ I suppose then I see an inconsistency since you agree that they are not exact duplicates but wish for them to be closed as exact duplicates. This kind of renders that close reason meaningless to me then. $\endgroup$ – Michael McGowan Oct 23 '12 at 15:15
  • 1
    $\begingroup$ @Michael I see your point, in that the site itself uses the phrase "exact duplicate." These issues have (of course) been discussed on Meta SO. I found this one to be particularly relevant to our conversation: Are duplicates creating broken windows?. $\endgroup$ – whuber♦ Oct 23 '12 at 15:30
5
$\begingroup$

This strikes me as a good and productive conversation (+1 to everybody). I just want to offer another option for dealing with these situations, and that is to comment on the main question leaving a link to other questions with related information / material. (Of course, when someone votes to close as a duplicate, this is done automatically.) I'm not saying we need to always go this route, but it's a nice option to hold in mind.

This approach will offer many of the advantages listed by @whuber. Among other things, it will also help Google find the best answers on CV, because there will be more links pointing to the best threads.

Another issue is that I often find there are subtle differences in questions that are listed as duplicates, which can invite differing, but equally valid, answers. As a case in point, there was a question a while back where the OP said that they had taken a log transformation of a covariate to approximate normality, but found that they could no longer interpret it, and asked how to interpret a log transformed variable. This was merged with another question on how to interpret log transformed variables, but before it was merged, I had answered that the distribution of covariates didn't matter. My answer no longer made any sense in the new context, so I deleted it. (For the record, I'm not complaining here, I'm perfectly fine with the questions having been merged.) Should a case like this exist, commenting and linking is another option that can be considered.

With respect to the thread in question, I had voted not to close, and later to reopen. However, I'm open to the idea that I'm wrong on that one. I have also noticed times in the past when I voted to close a question, but others voted to keep them open. I suspect it is only natural that there will be cases where opinions vary.

$\endgroup$
  • 1
    $\begingroup$ (1) I just want to point out that although your answer suggests otherwise, closed questions come up in Web searches. right now, the first hit when Googling "How to get an introductory understanding of Machine Learning" is the (currently closed) question we are discussing. (2) Your point about how merges can sometimes make nonsense of existing answers is good. Many merges ultimately are compromises between retaining good existing answers (and putting them in an appropriate place) and possibly ruining some other answers. Some intermediate mechanism between merging and closing would be desirable. $\endgroup$ – whuber♦ Oct 23 '12 at 15:05
  • 1
    $\begingroup$ @whuber, you raise a good point re #1, although what I was getting at was the ability to find the other Q, not the closed 1; having more links pointing to it helps w/ Google-like search algorithms, & it also means there are more paths that people can stumble across. (Again, I recognize that when it's closed as a duplicate, a link is given automatically & that this attenuates the force of my point.) It's worth pointing out that if a thread does have some merits, it can still be found & accessed when closed, but nb the fact of being closed may also serve to frame the thread in a poorer light. $\endgroup$ – gung♦ Oct 23 '12 at 16:05
  • 2
    $\begingroup$ In response to a separate comment thread with @Michael McGowan, I am coming around to favoring an idea suggested on Meta SO, which is to edit near-duplicate questions by including a prominent link to related threads at the end of the question itself (not just in comments). It's a bit more work, and ought to be supplemented by inserting reciprocal back-links in the original questions, too, but might go some way towards addressing the concerns raised by Douglas Zare. $\endgroup$ – whuber♦ Oct 23 '12 at 16:09
  • $\begingroup$ That's a good idea. I've been doing essentially that, but only in comments. Doing it in the body of the Q will make it more salient & allow you to add some context to the links that may not fit as well otherwise. $\endgroup$ – gung♦ Oct 23 '12 at 16:12
  • 2
    $\begingroup$ The specific recommendation is to terminate the question with a third-level header (i.e., preceded by "###" in markup), such as "###Related Questions", followed by links in the form "[question title](URL)". Low-rep users may continue to suggest such cross-references in comments (as many already do, quite helpfully) and higher-rep users can use these suggestions--as well as their own inclinations--to make such edits. $\endgroup$ – whuber♦ Oct 23 '12 at 16:23
  • $\begingroup$ As an example, I edited the question under discussion, voted to re-open it :-), and placed a back-link in the older related question. $\endgroup$ – whuber♦ Oct 23 '12 at 16:25
3
$\begingroup$

I voted to re-open. That's my tendency.

When a site serves several purposes (as this one does) some actions that are best suited for one purpose will be ill suited for another. One function of this site is to be a sort of encyclopedia of statistics answers - although, of course, there already is Encyclopedia of Statistical Sciences, many will not have access to it, it's expensive, and it isn't exactly what we are. There is, of course, also Google.

Another purpose is to the individual who asks the question.

To serve the first purpose, closing is natural. To serve the second, leaving things open is natural.

$\endgroup$

You must log in to answer this question.

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