I have a model built with time series data that I want to deploy. Sadly, I haven't found a lot of resources for this and I wanted to ask how to best approach this. The answer I am hoping for contains things related to:

  • Best practices
  • System design considerations
  • Possible pitfalls stemming from feeding truncated timeseries to a classifier trained on full timeseries

Would this be on-topic? If not - where would I ask this? I kind of assumed that SO would be more about implementation details instead of the big picture I am concerned about.

PS: I was also unsure whether it is rude to ask on meta if I should ask on CV proper or if it rude to ask on CV without being sure that my question is on-topic.

  • 8
    $\begingroup$ You're not being rude. The difficulty is knowing precisely what kind of question(s) you have in mind. It sounds as if you're wanting to read a discursive yet also detailed tutorial or review, not a concise answer to a precise, specific question. Best practice in deploying time series models seems likely to run some risk of being voted to close, as being too broad, or unclear what you're asking, etc. Being on-topic usually means that you can find on CV that similar questions were answered and the question and answer(s) were upvoted. $\endgroup$ – Nick Cox Apr 15 '19 at 11:57
  • 1
    $\begingroup$ I would like them to on-topic because to my mind these are the actual applied statistics but I think most users would consider them off-topic. (And thus you get Data Science on occasion.) If the question is carefully worded it would pass as a somewhat methodological but then again... I have little hope that is will be answered. In fairness, I do not think any SE community adequately caters these questions. You might get one or two gold nuggets. Operational research people might be better listeners. $\endgroup$ – usεr11852 Apr 15 '19 at 17:39
  • 6
    $\begingroup$ I agree with @usεr11852 Consider this: people able and inclined to write papers or textbooks are much more likely to write papers and textbooks than lengthy, detailed answers here. Many of us regard SE as some mix of (1) fun (2) instructive to us too (3) a kind of public service (the order of importance of those varies personally) but there is zero financial benefit and minimal career benefit to participation, which constrains the kinds of answers likely. $\endgroup$ – Nick Cox Apr 15 '19 at 17:56
  • 1
    $\begingroup$ All that said, you could try a fairly short question and see what happens. If nothing much positive, we did fear so. If something positive, well and good. $\endgroup$ – Nick Cox Apr 15 '19 at 17:58
  • 2
    $\begingroup$ Difficult to answer without the actual questions but of the three you quote the last seems most likely to be on-topic so if you want to test the water why not start with that one? $\endgroup$ – mdewey Apr 16 '19 at 12:54
  • $\begingroup$ stats.stackexchange.com/questions/404687/… this could be an example of such question. $\endgroup$ – Jan Kukacka Apr 24 '19 at 12:11
  • $\begingroup$ stats.stackexchange.com/questions/406173/… is another example $\endgroup$ – Jan Kukacka May 2 '19 at 7:38

You must log in to answer this question.

Browse other questions tagged .