Similar to W&B?

[fgolemo]

Hey Fabien,

Cool library but couldn't you use Weights&Biases for the same purpose?
I'm not sponsored by them or anything, but I have a lot of successful projects that use their libraries to do pretty much the same - create a snapshot of the code, parameters, machine stats, etc. But with the bonus of also hosting them for you and showing you plots.
So I'm just wondering, is there something neat that this library can do, that W&B can't?

Cheers,
Flo

[benureau]

Hi Florian,

Great question :)

Weight&Biases seems great. It seems tailored for ML and probably does an amazing job at it, whereas the reproducible library is quite agnostic about what you are doing.

It's also purely local, no account needed. Which means that if someone want to run my code, they don't have to create an account somewhere or edit part of the code. It also means that you keep the functionality whether W&B keeps existing or is bought by another company.

The code is simple—it's only one file, 500 lines of code at the moment. It's open-source, forkable. I only found W&B client code online, I might have missed the server and visualization part. And of course reproducible is completely free for everyone, forever.

I am glad that company are starting to jump into providing tools that enhance reproducibility. We need different tools with different usercases, skill-levels and degree of support.

Best,
Fabien