Best practice to compare the results before and after code change

[liutiming]

Hi @wlandau , I remember our discussion of data version control and was thinking it might be relevant to this. I often get into situations where I want to compare the results before and after code change. I am wondering if you or anyone has a workflow for that?

[liutiming]

For reference, https://dvc.org/doc/start/data-versioning is sufficient for toggle between project status, but I have not found a way to compare without significant code duplication

[wlandau]

Thanks, looks pretty slick to use dvc ... for _targets/ and git ... for code. Would you elaborate on the code duplication issue?

[liutiming]

Yea it is taking quite long for a large _targets though...

RE code duplication, if I have analysis a -> z, and I realise a -> b has a bug so I want to correct that. But I also want to see what was the impact of the bug on z, so in these cases I have to duplicate a2->z2 to compare z to z2 directly in targets.

One non-reproducible way to do it is to save z to a file and then compare that.

[wlandau]

Git branches solve the code duplication problem quite nicely, as long as they have informative names. Extensions to Git such as VSCode Git Graph can help manage branches locally.

[liutiming]

thanks I will try it out!