A new user asked this question:

Neural Network for Foreign Exchange Prediction

In my opinion it's quite good as a first question, but it's hard to answer. I'd like the help the OP improve it. However, I'm not sure how. The main problem IMO is that we can't say why her/his code doesn't obtain the same results as the book's code, because we don't have access to the book code. I was thinking to suggest two improvements:

  1. Modify the question in "did I make any errors in my implementation of this NN for Foreign Exchange prediction?" and add code, interspersed with text so that the post doesn't become an unreadable wall of code
  2. Add a link to the data, and some plots

However, at this point it seems to me that the question will become more on-topic on Stack Overflow than here. So I thought to suggest splitting the question in two questions: one, on topic here, where the OP describes the model more in detail, plots the data, etc. and asks if s/he is making shuts mistakes in the model definition. Another, on topic on SO, where the OP shows the actual implementation of the model. It's a lot of work for a beginner, but I don't have other ideas. Directly posting the code to SO doesn't seem right: I'm now due that talking about Hampel filters, AR coefficients, etc would make the question on topic there.

What do you think?

  • 4
    $\begingroup$ Not my field, but that question looks doomed to me. Why does my code which I am not showing produce results that puzzle me? is a question most of us face from time to time but it can't usually be answered remotely. But on your Meta point you can certainly make suggestions about changing the question but most of the kinds of suggestions that can be made would make this a difficult question for SO, not here. $\endgroup$ – Nick Cox May 3 '18 at 8:00
  • $\begingroup$ @NickCox sorry I don't understand. If I suggest the OP to include code, do you agree this would make the question less suitable for CV? $\endgroup$ – DeltaIV May 3 '18 at 8:10
  • 3
    $\begingroup$ Absolutely. An answerable question would need to show more details about what was done. It's possible that would reveal statistical or machine learning issues but more likely that it would morph into a question about the OP's code. I am active on SO and positive about it but even SO doesn't work well with fuzzy questions about complicated code. $\endgroup$ – Nick Cox May 3 '18 at 8:14
  • 1
    $\begingroup$ Since it's meta -- I agree to the observation of splitting the question and posting it to the correct sites. But how to hint gently at this approach? Looking at the topic of the question, my guess is that it's somebody learning to use NN. But a well-reviewed and completely available example will be better suited; the so called "detailed description" is not nearly detailed enough. $\endgroup$ – cherub May 4 '18 at 13:50
  • $\begingroup$ @cherub I agree with you. I'm starting to think that the question is doomed, as Nick Cox predicted. $\endgroup$ – DeltaIV May 4 '18 at 17:27
  • 3
    $\begingroup$ +1 for trying to help and improve $\endgroup$ – Sextus Empiricus May 7 '18 at 8:27
  • $\begingroup$ thanks @MartijnWeterings $\endgroup$ – DeltaIV May 7 '18 at 9:25

You must log in to answer this question.

Browse other questions tagged .