Dvc: Add DVC repro --recursive

[prihoda]

It would be nice to be able to use dvc repro -R on a directory just like dvc pull -R. Hopefully it won't be confusing that dependencies found outside this folder would also be reproduced.

[ghost]

@prihoda , can you explain your use case for this?
I'm imagining that you have a directory with some stage files, and you want to reproduce only the stage files that are found in that directory. If so, how did you arrange your project?

[prihoda]

For example, a directory can contain results of one separate data analysis and I want to reproduce them. I know I can create a Dvcfile with all the outputs but it would need to be updated each time I add an output figure.

[ghost]

@prihoda , are the stage files on such directory related? How does the pipeline looks like?
Sorry for bordering you with a lot of questions, just trying to understand better the scenario before implementing this feature :sweat_smile:

[prihoda]

Sure I understand 😄It's the same usecase as dvc push -R and dvc pull -R really. The stages are related for the same reason that they are put in the given directory 😄For example, it can be ROC plots and Precision-Recall plots for a set of trained models.

Directory structure:

data/
    models/
    plots/
        rocs/
        pr/

I would like to run dvc repro -R data/plots/.

[prihoda]

Working on it https://github.com/iterative/dvc/pull/2030

[ghost]

Thanks, @prihoda !

[jorgeorpinel]

I was just passing by and noticed this issue is still open. Should it have been closed along with PR #2030 ?

[pared]

Good point @jorgeorpinel!