Juan M. Cruz-Martinez

The way Keras used to do this was with environment variables so that at first instantiation you get the one backend and that's it. I'm happy with this method as...

Rebased from master to get the fix for travis. So in the end what do you guys prefer, a environment variable that selects the backend (and then by default you...

kept as a draft for now until a third backend is added

@RoyStegeman if you don't mind I'll keep rebasing this branch periodically to the current master because I want to keep it up to date.

> @scarlehoff do you still have plans for this PR? I think the other backend you were considering was for the quantum computer project, but apparently that didn't end up...

It is a working version so if someone is really interested they can run fits with `n3fit` with a GA.

The machinery is needed for a more general "choose your backend" scenario. I would leave it open because maybe in a few years time it is useful to some unsuspecting...

It was my first thought then I saw we have O(3000) fits in the server so I don't know how big that json will be or whether it will be...

I guess we can ask the INFN for a bit more juice if we need.

Just a note to myself: this can be also exploited to search by theory config, which would allow the download of theory covmats for which we already have fits.