# Can I ask questions about an R package for machine learning with unclear documentation?

Machine Learning (in general) is on topic (see: Are the "Machine Learning" questions on topic?).

I am working with the h2o package and the documentation is .. obtuse. It is hard to use and generally uninformative. I want to ask (and maybe answer) questions that document things like "here is the code to upload a text file to h2o from your interface" or "here is how you start the evaluation of data using a pre-loaded and pre-built rf/glm/gbm/dl/nn/... ".

Can you tell me if doing this is good or bad, or if I should ask these questions elsewhere, like in Stack Overflow?

• This post appears to have tailed off mid-sentence! (I've seen this happen to a couple of people before, and I wonder what causes it.) Feb 5 '15 at 1:59
• The generic advice at stats.stackexchange.com/help/on-topic on software-related questions seems to apply here. Else in what sense is that advice problematic or irrelevant? Feb 5 '15 at 12:31
• It is the half-and-half world between programming (details of the language) and the fundamental idea behind the language - aka the Machine Learning part. The answer that I referenced seems to say a loud yes. The on-topic generic advice seems to say "get the heck out". I'm not sure which to listen to and if you want one then we as a community don't get the other. Feb 5 '15 at 14:10
• Think there's a difference, though, between the examples from the answer you referenced - e.g. "Why does the kernelized support vector machine algorithm care that the kernel function be positive semi-definite? What happens if it isn't?" - & the examples you propose - e.g. "here is the code to upload a text file to h2o from your interface". The former don't sound at all like the "routine data processing" or "details of the language" referred to in the guide to what's on topic; the latter do. Feb 5 '15 at 16:19

If your only question / planned answer for the thread is "here is the code to upload a text file to h2o from your interface", then the question will be fine on Stack Overflow.

On the other hand, if you want to talk about "the evaluation of data", that is on-topic here, even if code is appended. There is certainly nothing wrong with having code (R, SAS, Stata, MATLAB, etc.) as part of a question or an answer. My opinion has always been that what is important is what the OP needs explained, regardless of how the question is phrased. If what the OP needs to know is about the machine learning, then it is best here, but if what they need to know is about the code, then it belongs on SO. (If they need both, then it could be either, but we can defer to the OP.)

Just try to make your questions real questions about machine learning, not primarily about code. Also note that we are a pure Q&A site; we don't really do tutorials. However, if you want to put a tutorial together, that would be a good candidate for a CV blog post. For more on the CV blog, see:

• where can I find out more about CV blog posts? Right now deep-learning is exploding and 'h2o' is decent package for it. In my looking, I have not found any "in the wild" tutorials. I think that if such things were associated with stackexchange they could substantially improve its value to the community. Feb 5 '15 at 16:29
• @EngrStudent The CV blog is here: stats.blogoverflow.com. If you want to contribute, you can join our Ten Fold chat room and we will create an account for you.
– chl
Feb 5 '15 at 17:14
• As someone active on SO too, I want to underline that many code questions are not appropriate on SO either. SO bills itself as for professional and enthusiast programmers and focuses on specific code-based questions. It doesn't really aim at serving people whose coding problem is "I need code and don't have any", although capriciously sometimes such questions get answered. All forums I know about have a tacit or explicit policy that you should Read The Fine Manual first, for all that increasing numbers think asking on the internet is the way to avoid reading documentation. Feb 5 '15 at 17:51
• Also note there's a Google Group dedicated to h2o. Feb 6 '15 at 11:35
• @NickCox - Sometimes the manuals are more easily used than others. I see this as potentially making a more easily used manual in a forums where people are looking for it. Feb 6 '15 at 16:52
• Twofold reaction: First, I am sympathetic. Outside this forum, a large fraction of what I do is expository and consists of going beyond the documentation for my favourite software (which is excellent!). Second, and more crucially, weak documentation is a problem for anybody who uses software -- in essence, anyone here -- but it's not obviously a problem for this forum. If people want better explanations of statistical operations, their questions will often (usually?) fit in here. If people want better explanation of the why or how of some software, then more often (more usually) not. Feb 6 '15 at 16:58