Too vague contribution guidelines

The contributing guidelines given everywhere tell only the development git workflow and does not actually say how to setup the development environment.

I think it should let new contributors know the packages required for installing scikit-learn, for example, numpy, scipy and cython.

[jnothman]

Well if it makes you happy, it can’t be that baaaad…

On 5 September 2017 at 09:33, Caio Oliveira [e-mail] wrote:

I agree, unfortunately. I wouldn’t remove manual install instructions, but providing an “easy path” with docker would be great. As I said previously, writing make devtools; devtools/env.sh make makes me really happy 😃

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/scikit-learn/scikit-learn/issues/9687#issuecomment-327039149, or mute the thread https://github.com/notifications/unsubscribe-auth/AAEz63mG_jxVaT_-RRYytgK-yAVo59-fks5sfIjNgaJpZM4PMJdC .

[jnothman]

Never mind docker. Anaconda goes a long way to not needing Docker for something like scikit-learn. At the moment our docs talk about a full Anaconda setup, or using conda to install scikit-learn. What we should do in the contributor docs is describe how to use conda to set up an environment, install dependencies and then do an in-place build of scikit-learn for development. We could also consider an environment.yml to make this “one-click” with conda-env.

On 14 September 2017 at 22:29, Roman Yurchak [e-mail] wrote:

(from @caioaao https://github.com/caioaao 's #8960 (comment) https://github.com/scikit-learn/scikit-learn/pull/8960#issuecomment-327026652 ) we could generate images that mirror what CIs are running, make everything reproducible with minimal effort, etc. I did this after a merge from master that broke the build on my machine. python users usually can rely on virtualenv/anaconda, but gets messy really fast, specially when you’re the developer (in contrast with being the user of the library). being able to run make devtools; devtools/env.sh make is the optimal flow for me: this way we only need docker installed and nothing else.

The idea of getting full reproducibility of CIs is interesting but IMO it’s more the advanced usage and not something a typical person would need to start contributing to scikit-learn. Just conda virtualenv should be enough in most cases; I don’t see why it would get messy (one can always re-make a clean env in case of doubt). As @lesteve https://github.com/lesteve said, adding docker to the official instructions would mean that somebody would need to spend time supporting it (e.g. “I have Windows XP and my docker container with scikit-learn dev is not working as it should…”). However, having an official docker setup for sklearn dev environement would be useful (maybe under a separate repo or in some subfolder of scikit-learn/scikit-learn) so that it would not need to be re-invented each time?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/scikit-learn/scikit-learn/issues/9687#issuecomment-329466734, or mute the thread https://github.com/notifications/unsubscribe-auth/AAEz6zPrhcUh-j66Nd-44TDFf3cHO27Uks5siRw3gaJpZM4PMJdC .