At which point do implementation questions become off-topic?

Question

I haven't ever posted here, but I've been intermittently lurking for some time, in order to understand which kind of questions get asked here. In particular, I see that implementation questions are off-topic, but I've seen a few implementation-related questions such as this one:

RNN LSTM not converging with Adam

EDIT: I rephrase my question to avoid misunderstandings. Your help pages says that questions about "[..]the implementation of machine learning" are off-topic here, however it seems to me that a few questions with code are accepted (but I may be wrong! Please let me know if this is the case). So, do you accept questions containing code, as long as they are not exceedingly focused on implementation details?

Answer 1

Speaking as a mod, I tend to use a light touch in regard to migrating many of these potentially out-of-scope questions. (Similar to avoiding closing questions by fiat, in all but the most egregious, preferring to follow the will of the community.)