10
$\begingroup$

I posted a question on here recently that was closed:

My closed machine learning question.

I was unsure if it fit into this site anyway, so I'm fine with it being closed. But it left me wondering, what is an acceptable Machine Learning question on Cross Validated? The on-topic section of this site says Machine Learning is acceptable:

Acceptable topics on Cross Validated.

But it's not clear to me what aspects of Machine Learning you're allowed to ask about. Can someone provide clarification?

$\endgroup$
9
  • 7
    $\begingroup$ One test we often use for questions is "Does this question require statistical knowledge to answer?" But this question could be summarized as "How do three different software libraries compare and contrast", which seems to be primarily about how the software works. (Also, it is an enormous question; I didn't vote to close, but if I did, it would have been for the "too broad" reason). Substantially narrower versions if this question could be on-topic, provided that the answers to those questions requires statistical expertise. $\endgroup$
    – Sycorax Mod
    Jan 6, 2023 at 14:24
  • 2
    $\begingroup$ If I narrowed it to: "These two algorithms are very similar. What's the main mathematical difference between them?". Would that be considered on topic? Or perhaps, "From a statistical standpoint, how does the difference between these two algorithms effect their results?". Or maybe, "Why doesn't XGBoost use histogram bagging for its features?". $\endgroup$
    – Connor
    Jan 6, 2023 at 15:07
  • 7
    $\begingroup$ The first two questions are rather open-ended, and would tend to invite unfocused answers that speak in glittering generalities. The last question suggests that the thing you want to know about is histogram bagging (which is good, it's crisp & statistical), but is framed as "Why didn't the XGBoost authors do [this specific thing]?" which is a question that only the authors could answer. Perhaps you want to know "What are the advantages and disadvantages to histogram bagging in gradient boosting?"? That would be a good question. $\endgroup$
    – Sycorax Mod
    Jan 6, 2023 at 15:28
  • $\begingroup$ Thank you, that clarifies everything apart from the statistical part of "crisp and statistical". What do you mean when you say "statistical"? Does "statistical" mean any area of maths or application that falls within the theory of statistics? How can I assess my question's level of "statisticalness"? $\endgroup$
    – Connor
    Jan 6, 2023 at 15:34
  • 2
    $\begingroup$ In that usage, I was trying to underscore the difference between "Why did the XGBoost authors make this design choice regarding histogram bagging?" (not statistical -- it's about the authors) and "What are the pros and cons of histogram bagging?" (which is statistical, because it's about histogram bagging). $\endgroup$
    – Sycorax Mod
    Jan 6, 2023 at 15:35
  • 1
    $\begingroup$ Okay, but the idea of something being "statistical" or not has come up in the answer to this question to. What does "statistical" mean to the mods of this site? I'm not asking to be clever about the definition, I'd like to know so I can correctly evaluate whether or not to put a question on here. It seems to me like Cross Validated is for theory, and Data Science is for practical applications of theory and other practical matters, do I have that right? $\endgroup$
    – Connor
    Jan 6, 2023 at 15:39
  • 5
    $\begingroup$ I think the help center does a good job of articulating the scope of the site, and the dictionary meaning of "statistics" seems sufficient. I don't agree with your characterization of the two sites' scope, because statistics is a practical discipline. I'm not an expert on what is on-topic at Data Science, so I don't feel that I can give a complete answer describing how the sites are different. $\endgroup$
    – Sycorax Mod
    Jan 6, 2023 at 15:44
  • 4
    $\begingroup$ As a cursory example, DS seems to accept questions which are about how to do tasks on a computer. This question asks how to use a certain library on a computer not connected to a network. datascience.stackexchange.com/questions/117575 This question asks how to concatenate two CSV files. datascience.stackexchange.com/questions/62934 A person might say these are statistical, because modern statistics happens on computers, or because CSVs store data, or statisticians use software, but CV views each as incidental to statistics itself. $\endgroup$
    – Sycorax Mod
    Jan 6, 2023 at 15:49
  • 5
    $\begingroup$ It is perfectly fine, btw, for Data Science to have a different scope, and this is not intended as a slight in any way. "Data Scientist" is a job role that often includes statistics as one knowledge area. (But not always -- I've worked at organizations where the essential skills for a "data scientist" were the ability to write SQL queries, create pivot tables in Excel from the results, and then put that into a PowerPoint slide.) Many jobs require several different skill sets, and DS.SE seems scoped to comprehensively address the "data scientist" job role. $\endgroup$
    – Sycorax Mod
    Jan 6, 2023 at 15:56

1 Answer 1

7
$\begingroup$

Re , its acceptability had been unanimously discussed in ML questions: here or at Data Science? and Is machine learning a part of statistical analysis?.

The scope of the site is crystal clear: anything pertaining to statistical concepts that OP listed above.

Coming to the present context as to why the question was closed - the reason has been mentioned:

This question appears to be off-topic because it is not about probability, statistics, machine learning, data analysis, data mining, or data visualization.

That is, the close voters couldn't churn out any statistical query in it. In fact, it was asking to compare two algorithms, their speed etc. Now, asking about algorithms is not forbidden provided there has to be a specific and explicit statistical query about it, which in this case, seemed to be at best not clear or outright absent.

In fact, it is further mentioned:

In edge cases or for very opaque questions, it can be helpful to ask what OP's statistical content of the question is, which might prompt OP to revise.

It might be that the close voters failed to comprehend any legit statistical query that OP tried to articulate. So, edit the post to clear it up what the relevant statistical content was!

Closing is not the end of the day; rather it's an opportunity to reframe the post so that it might be more comprehensible. It would go to the Reopen for the community to vet whether it was modified to an extent enough to reopen.

$\endgroup$
5
  • 6
    $\begingroup$ This (+1). Plus consider that sometimes a question can seem unsuitable on several grounds, but votes to close can only mention one reason. I wasn't involved in this case, but I would have voted to close. The question looks enormous; indeed, there's more than one question. Please try some role reversal: how long would it take an expert to answer this question? how much would you expect an expert to write? My own rule of thumb is that a question I know something about might take me 5-10 minutes to answer if I am lucky. I might choose to spend more, but an OP should not expect more effort. $\endgroup$
    – Nick Cox
    Jan 6, 2023 at 13:12
  • 1
    $\begingroup$ 1. That's fine, the question was placed somewhat speculatively! Is the Data Science stack exchange a better place to ask something like this? 2. @NickCox I agree, it's a huge question! I posted it to see if anyone had an answer to hand, however, my intent is typically to answer it myself if no one else does. 3. Could you define what the question's "statistical content" means? For example, if I asked about the difference between the mathematics of the algorithms, would that be considered a statistical question? Does Cross Validated lean towards theory? $\endgroup$
    – Connor
    Jan 6, 2023 at 14:01
  • 5
    $\begingroup$ I will answer #2. It's human beings making decisions about your posts, so there is some caprice about how you get answered. If some expert had seen your question and thought "I can answer this" you would have been lucky. Just occasionally here there is a question which I understand thoroughly, have studied in detail myself, and have already written about. That is the luck you need. Meanwhile, you've deleted the post but if it was still up many people have enough reputation to vote to reopen. But as said I can't think, on the evidence, that the decision to close was wrong. $\endgroup$
    – Nick Cox
    Jan 6, 2023 at 14:07
  • 2
    $\begingroup$ I don't disagree with the decision at all, it's fair criticism, thank you for your insight. $\endgroup$
    – Connor
    Jan 6, 2023 at 15:05
  • 2
    $\begingroup$ This is saying that only statistical questions about machine learning are on topic here, which seems like a good idea but not the policy officially stated in the help center. (Applying the official policy to this question led to a close reason saying “it is not about…machine learning”, which is false.) I don’t think answers, close reasons, and help text written in 2010 can be a good guide on scope issues around machine learning in 2023, since that field has grown so much. $\endgroup$
    – Matt F.
    Jan 19, 2023 at 11:09

You must log in to answer this question.

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