-8
$\begingroup$

I respect the moderators in general but this time their behaviour is just super weird to me. Please read these 2 links nn.embedding similar general purpose for float values how nn.Embedding algorithm makes embedding vectors and I think someone not with a lot of prior knowledge can judge who is right?! I clarified my position in comments that you can read, but to summarize:

  • all I want a guidance to improve my question but moderators don't let proper people reach me!!!.
  • You can see my efforts to clarify my question in my question edits and in my comments trying to follow moderators instructions, trying to come up with new details and explain whatever I can, like in my 1st post I edited it a lot adding useful details, made 2nd post more clear without insisting to reopen my first post again and trying to avoid argument, so all these show I'm not here to argue.
  • ofc for a question like mine there are lot of ways to help me but one of possible one could be "For discrete data there is a method named "x" and for continuous data I'm asking if someone have a ready code for method "x" in which to assimilate discretion, partition float data to periods and puts centroid of period data as a representation of that period (viola now we have a discrete like data) then uses method "x", so please tell me how method "x" could be implemented, do u have useful sources and show me where a code like that is? btw I'm sure you know other ways of discretion of data also exists."
  • Their reason is that "you don't know this method" and my answer is "isn't that why everyone asks a question".
  • They don't let my question to be open in the case if it really needs clarification, I would get what I should add while these moderators just enforce themselves meddling into my question not asserting "what should I add" while Ive done whatever I could.
  • Taking away chance to get an answer while it's obvious that someone here most likely know the answer, like on my other post related to this in less than 10 minute and 1.5hour I got two good answers, so again moderators insist on closure of this post is suspicious, because if someone has proper knowledge on this issue they can ask me their needs of clarifications, and I would gladly answer them.
  • They are unable to answer "what should I add?", "what I should focus more on", "why u don't let if anyone knows that help me" in countless cases, making these whole arguments not productive, they just want it closed!; all I want is please tell me what to do, they said more precise and I did whatever I could but when they r unable to convey what they want, it just senseless no matter in what position you are.
  • I admit sometimes in these arguments I was a little bit sarcastic but in my defence just follow places of such, it's always after their inabilities of clarifying their intention on insisting on it unreasonably.
  • to be brutally honest its a bit awkward people with such sophistications can't respond to a normal question like "what I should add more" etc and insist on their decision and while trying not to be pessimistic or accuse them cause maybe I'm wrong but it just seems they re just defending or try clean up the fault of their co-moderators

I think a person with common sense knows who should give right to. but I'm open to hear your opinions. please tell me what should I do? What do you think?

after many people coming and checking this discussion and just upvoting my opposite opinions and downvoting me, literally no one giving the right to while I have made several points to open the my question, answering that Ive addressed their points as possible in the comment of @Tim 's answer. I'm just thinking either I'm wrong in the all of my logical cells in my brain don't help or then again the last bullet Ive mentioned here, cause people still don't answer my questions and they just support their mates. but at least a single person, I don't say fully supports me, but realizes parts of my reasons, would have been a heartwarmin.

if u want to be productive please just tell me about why my bullet no.3 can't or may can't be implemented? I'd like to listen just about this, if u want to make a reasonable closing reason.

$\endgroup$
19
  • 5
    $\begingroup$ I haven't had a chance to investigate yet, but if your question is asking for code, that is a reason for closing. $\endgroup$ Commented Oct 2, 2021 at 12:31
  • $\begingroup$ but their reason for closing my q is that "I don't know what I want?"!!!! lots of possible solutions are there to help me and it's not only about me asking for code beside if then have a code that would be a lot better. please please read these 2 links and tell me ur opinions. they(answerers) can provide details how to do method easier myself, lots and lots of ways of helping as I said. $\endgroup$ Commented Oct 2, 2021 at 12:40
  • 7
    $\begingroup$ You're asking for views and here's mine. I didn't vote or comment on either of the original threads but I now consider both as suitably closed at present, on any of several grounds, principally the extent to which there is seemingly a request for code and that each question is written unclearly and without a simple focus. It strikes me that different moderators have tried patiently to deal with your points. The way is to move forward is to rewrite your questions, not allege unfair treatment by moderators who are doing what they are supposed to do, and doing it very well. $\endgroup$
    – Nick Cox
    Commented Oct 2, 2021 at 13:13
  • 8
    $\begingroup$ A thread has to be based on a clear and specific question. If you don't have one and you just want to start a broad discussion then (a) that's risky because the question is likely to be closed as too broad or lacking focus (b) no one else is obliged to try to guess what your real question is, or should be. $\endgroup$
    – Nick Cox
    Commented Oct 2, 2021 at 13:23
  • $\begingroup$ how broad do u think my question is? it's focused on specifically nn.embedding trying to understand underlying code better, trying to how to get a variant of it, if with code better. then u r saying "lack of focus?!", please specifically tell me about lack of focus in these two. $\endgroup$ Commented Oct 2, 2021 at 13:28
  • $\begingroup$ uncountable number of questions are there lacking much lot focus, asking to explain something better is one the key types of questions here $\endgroup$ Commented Oct 2, 2021 at 13:31
  • $\begingroup$ I suppose all your sayings are correct, so let my question be open for a while, if someone know more about this topic comes and ask me more precise questions ,then I have explain him what I want, throwing some "lack of focus","need to add details" without addressing it, is not helping, respects $\endgroup$ Commented Oct 2, 2021 at 13:37
  • 7
    $\begingroup$ I can't be surprised that you disagree with my view, But you're raising the same points again and again. The burden is on you to withdraw or rewrite your questions until the community accepts them. That's how CV -- and SE generally -- works. Also, the existence of other poor questions is neither here nor there, as it's your threads that are problematic. $\endgroup$
    – Nick Cox
    Commented Oct 2, 2021 at 13:46
  • $\begingroup$ I did the thing u say at second try trying not get into debate, providing more detail, stating it more clear, but then again no one answers my questions, u just label it some way and when u r asked to address it again don't write about anything related. $\endgroup$ Commented Oct 2, 2021 at 13:50
  • $\begingroup$ what could I have done to be more clear? when u don't address what details I should add, why u don't answer it? when u don't know what should I add why u insist on ur opinion? $\endgroup$ Commented Oct 2, 2021 at 13:53
  • $\begingroup$ I had a question about this topic and they answered it less than half of an hour, stats.stackexchange.com/questions/543324/… but you without knowing anything seems just trying backing up ur comoderators, and I ask again please say what should I add? $\endgroup$ Commented Oct 2, 2021 at 13:59
  • 8
    $\begingroup$ I am bowing out here. You asked for opinions. You got one, mine. The issue here on Meta is Have you been badly treated by the moderators? and my answer is No, not on the evidence I see Being disappointed at the closure of threads is one thing. Showing anger and making personal comments won't help your case, (I am not a moderator, as my profile makes clear.) $\endgroup$
    – Nick Cox
    Commented Oct 2, 2021 at 14:00
  • $\begingroup$ I usually don't talk like this, I know im a bit harsh, bit when I feel u r not being fair just because we r not in a same teams it messes me up. ofc I also understood u didn't know the answer to "what should I add", had no answer to "why u don't let if anyone knows that help me", "what I should focus more on" $\endgroup$ Commented Oct 2, 2021 at 14:05
  • $\begingroup$ do u think I like to just repeat my 10 reasons again and again? believe me, im not, but when I just here someone labeling my question with something without saying where that label matches ofc its frustrating for me. $\endgroup$ Commented Oct 2, 2021 at 14:08
  • 2
    $\begingroup$ That was and remains a factual correction. Your comment could easily be read as implying that as another moderator I am unsurprisingly agreeing with the moderators concerned. But as said I am not a moderator. My comments here rest on what I read but do draw on my experience on the site. If there are other views, people will add them. $\endgroup$
    – Nick Cox
    Commented Oct 2, 2021 at 14:19

3 Answers 3

14
$\begingroup$

In case of both questions you asked me and @Sycorax had a lengthy discussion with you in comments.

  • We both suggested that you should make the question more precise because it is not clear what you mean by "embeddings for float variables". In neither of the cases, you did anything to edit the questions and clarify them. We gave you guidance on what can be improved.

  • We both took attempts to clarify it and partially answering it in the comments by pointing out that the equivalent of embeddings for continuous variables is using a dense layer since embeddings are dense layers, just the fact that we are dealing with sparse, binary vectors enables us to use a more efficient implementation that the direct matrix multiplication.

  • Consider that two of your questions were closed by different moderators. In this thread, you got comments from other users commenting in favor of the moderator's decision. You asked exactly the same question on another SE site and got no response and no upvotes. It seems like there are at least several people for whom your question was not clear enough to be answerable.

  • Your question is basically an XY problem, you are asking about an attempted solution rather than your actual problem. You want to get answered on how to implement "embeddings for floating-point variables", but you do not clarify what you want to achieve by this and what kind of problem you are trying to solve, what is this going to do. The problem is that the thing you are asking doesn't exist, so we cannot help you with your intended solution. That is why we tried to clarify what is the underlying problem and what you mean by those "embeddings" but you failed to address the comments.

  • For further guidance, you can check our How to ask a "good" question on CrossValidated? guide.

$\endgroup$
3
  • $\begingroup$ in order of ur bullets:1."what you mean by 'embeddings for float variables'": one implementation on that may be my bullet 3 here |2."using a dense layer":answered here stats.stackexchange.com/questions/546775/… 3."got no response [there]":my exprience shows its more likely asking about neural nets don't get answers and people interested to neural nets usually hang around here, likewise u and syxxxx didn't know I have a question there to close it :/ :p $\endgroup$ Commented Oct 2, 2021 at 23:19
  • $\begingroup$ 4. "you do not clarify what you want to achieve by this and what kind of problem you are trying to solve": I have detailed method I want to implement through post and also here stats.stackexchange.com/questions/546430/… if Ive missed sth comment, and also kinda read "How to ask a "good" question on CrossValidated?" rn :/ was no help but I should read it later $\endgroup$ Commented Oct 2, 2021 at 23:19
  • $\begingroup$ I think have followed ur instructions or noted ,answered them somehow, at least I tried to. $\endgroup$ Commented Oct 2, 2021 at 23:25
6
$\begingroup$

if u want to be productive please just tell me about why my bullet no.3 can't or may can't be implemented?

The point number 3 is about a request for code implementation.

For discrete data there is a method named "x" and for continuous data I'm asking if someone have a ready code for method "x" in which to...

(emphasis mine)

This is extremely off-topic here. The website is about questions about 'statistics' and not about 'how to write code' or 'where to find code'.

https://stats.stackexchange.com/help/on-topic

if it needs statistical expertise to understand or answer, ask it here; if it's about the implementation of an algorithm, routine data processing, or details of the language, then please refer to the collection of links to resources we maintain.

Of course, code still occurs on the website. But only as a means to improve a question about statistics or when it statistics oriented. But 'ways to implement an algorithm' are not about that.

$\endgroup$
8
  • $\begingroup$ 1."The point number 3 is about a request for code implementation.":no its obviously is requests to give resources to me to get to know how to do that, beside Ive appreciated the code also. is that clear? 2."tell me about why my bullet no.3 can't or may can't be implemented?": this means I say all readers of my question yet haven't added any useful suggestions(ofc except dense layer which is not help to satisfy for replicating transformer for float data cause it is just not truly cloning embeddings) so im looking for possibility of implementing such, or in other words, is that possible to do? $\endgroup$ Commented Oct 6, 2021 at 4:26
  • $\begingroup$ 3."not about 'how to write code'": isn't looking for resources for 'how to do algorithm' is on topics of this site? 4. I don't know how to do algorithm nothing code related,where should I ask it for resources and insights? $\endgroup$ Commented Oct 6, 2021 at 4:26
  • 4
    $\begingroup$ @Farhang The question is not about a topic that I know much about so it is difficult for me to comment much more about it. But... plainly asking for a code that does some algorithm only in some different setting than what is currently available (without clear significance what the value is for statistics) or resources about it is a gray area.... $\endgroup$ Commented Oct 6, 2021 at 8:06
  • $\begingroup$ ... The format of Q&A here prefers quick twitter style with simple questions and simple answers (in terms of size, the concepts might be complex). The way that you ask your question is heavily making use of python code. At least for people that are not directly into the field it is hard to get what you are doing. Sure there might be specialist that could possibly get what you are trying to do, but a good standard would be to write your question to people as to a broader public. Stack Exchange has a broad public and is not like a specialist journal... $\endgroup$ Commented Oct 6, 2021 at 8:08
  • $\begingroup$ ... again I am not a specialist and have no idea what 'embedding' means. And it seams like you want to embed floating numbers. That seems like a simple question. Why not refere to some journal article instead of all this python code and referring to error messages (which maybe obscures the point about statistics that could be underlying in your question)? ... $\endgroup$ Commented Oct 6, 2021 at 8:13
  • 1
    $\begingroup$ ... For instance something like this "In this paper (add link) a description is given of a method for embedding words which is implemented in the python package torch. This solves problems like such and such. My problem is slightly different and is as such. Because of that I would like to do embedding with floating point variables. Would this be possible and are there descriptions solutions of this?" $\endgroup$ Commented Oct 6, 2021 at 8:16
  • $\begingroup$ first of all I get ur point to not adding code, but for concepts by experience, Ive learned people don't get mixedup by codes(ofc easy to read, non specific syntax codes) and my code was simple except one specific syntax which my whole question was about its underlying algorithm, so technically any answerer probably should know how the torch works(another example would be how the algorithm method a in package a1 is different than the algorithm in the method b in package b1) so for answering this u probably should know how to read syntaxes package a1,b1.but anyway thanx to reminding me $\endgroup$ Commented Oct 6, 2021 at 17:17
  • $\begingroup$ anthor thanx for ur last comment depicting how such question should be asked. $\endgroup$ Commented Oct 6, 2021 at 17:18
4
$\begingroup$

I have no knowledge of embedding and neural networks, but formatting the question is already helping a lot to clarify a question.

Here is an example with the following changes

  • place code at the end
  • use capitals
  • add section headings
  • turn the long paragraph into points
  • remove stuff about the error message
  • remove repetitive use of 'please' and 'tell me'
  • highlighted question
  • removed question about positional embedding to improve focus

As I understood from comments by others who know this stuff better there might also be some XY problem underlying the closure and embedding float numbers is something you do not do. To clarify why you are thinking about doing this at all is to explain why you have this problem. For instance "question: can I add up 1+1 = 3? answer: no, but why would you want to do that? reply: because my keyboards '2' is stuck." To avoid that start with what you are doing.


My question relates to pytorch code of transformers suitable for machine translation. (See the code at the end of the question)

One the first steps is embedding the word vectors. But for none word integer purposes, is it possible to also use nn.Embedding or something alike so to have float numbers?

Why does nn.Embedding only deal with words? What are alternative ways to embed floating numbers which are commonly used (e.g in non-machine-translation, non-integer tasks like most of neural nets, time series analysis, CNNs etc)?

I know float numbers don't work with nn.embedding. My question is when we are dealing with integers, someway we embed them but what is usually done equivalently when we are working with float numbers?

  • One of possible ways is to use fully connected layer but in machine translation they also can use indexes(in form of float) and map them to another float vector ,except and avoid using word embeddings or in case of general (nn.embedding).
  • They probably trying to preserve some features of language and so thats why they use word2vec embeddings rather than using vanilla fully connected layers.
  • But, in terms of using transformer models to get float inputs, we don't know what should be preserved
  • So my question is what the other researchers of this field have done?

Example code

import torch
import torch.nn as nn
class Encoder(nn.Module):
    def __init__(
        self,
        src_vocab_size,
        device,
        embed_size):
        super(Encoder, self).__init__()
        self.embed_size = embed_size
        self.device = device
        self.word_embedding = nn.Embedding(src_vocab_size, embed_size)
    def forward(self, x):
        out = self.word_embedding(x)
        print(out)
        
if __name__ == "__main__":
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    x = torch.tensor([[1, 5, 6, 4, 3, 9, 5, 2, 0], [1, 8, 7, 3, 4, 5, 6, 7, 2]]).to(device )
    src_vocab_size = 10
    embed_size=512
    model = Encoder(src_vocab_size,device,embed_size).to(device)
    out = model(x)
```
$\endgroup$
1
  • $\begingroup$ its really appreciate that you tried to clear up my question even though lots of accusations of not following the rules (asking for code, having a code in q's text etc )was thrown to my question. and in this answer by clearing attributes necessary for q like mine, and by providing how exactly those should been written in the question made a great contribution to this debate. $\endgroup$ Commented Oct 6, 2021 at 17:37

You must log in to answer this question.

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