Use binder for notebooks

[tpoisot]

mybinder.org lets people run jupyter notebooks from github repos in the cloud for free -- it might significantly decrease install issues for reviewers.

[benureau]

I have some experience with mybinder.org. The "regular" service seems to have been unreliable for quite some time, and only the beta service https://beta.mybinder.org/ seems to be working right now.

While having a version of the code as a mybinder notebook is certainly a good thing to disseminate the research to a broad audience, it is not necessarily the best way to create a robust method to make the code run over the medium or long-term.

mybinder notebooks should be encouraged, but a more regular and resilient way to install the software should probably be required.

[rougier]

@tpoisot Do you suggest author to prepare a notebook for each submission (when possible)?

[tpoisot]

At least offering this as a possibility. This way, if the reviewers don't feel like re-installing the entire environment, they can run the code in the cloud.

[khinsen]

There are good reasons for propsing this option, but there is also one good argument against it: reviewers using binder will no longer check the installation instructions.

[jzf2101]

What about repo2docker? People can confirm both that installation works and build images locally. crowdAI does this for their competition submissions:

https://github.com/crowdAI/crowdai-musical-genre-recognition-starter-kit/blob/master/Round2_packaging_guidelines.md

[khinsen]

It would be nice to do a real-life test with such an approach. Ask a few submitting authors if they are willing to play guinea pig and use repo2docker. Then see how reviewers react. Note that verifying the local installation is a task mostly unrelated to the rest of the review, and thus could be delegated to someone else, even someone with a different scientific background (or none at all).