Data pipeline using DVC

[ijlyttle]

Demo of DVC with Quarto

DVC is an open-source project, an acronym for "data version control". It can be used to:

• specify the dependencies and outputs of a pipeline
• manage remote-storage of data

This sounds a lot like targets (for R) and pins (for R and Python), both of which I respect greatly. One of the big differences is that DVC can live outside your code, as an orchestrator, whereas, my understanding of targets and pins is that they are integrated into the pipeline code.

As an orchestrator, I wanted to try out DVC as a complementary tool to Quarto, so I made a small pipeline (repo). This pipeline is made using Python, but like Quarto, I don't think DVC "cares" about the languages used in the pipeline; it just runs quarto render on a file if it is invalidated. This required me to freeze the Quarto files in the pipeline, so that a global quarto render will work for the "rest" of the project.

I use an S3 bucket as backing, but DVC supports a variety of remote storage.

One point that I wanted to bring up here is that I found .qmd files to be suitable for the pipeline, but not .ipynb files. I may be putting this imprecisely, but running a .qmd file does not change the .qmd file; however, running a .ipynb file changes it. If a pipeline's dependencies includes a .ipynb file, I can see things becoming circular.

That said, there is an issue at DVC where they discuss allowing a dependency to filter for the (in this case) Python code - but it seems to have stalled.

FWIW, I'll be at posit::conf() if anyone wants to discuss. Thanks!

[cderv]

Thanks for sharing @ijlyttle !

DVC is indeed an interesting tool ! I won't be physically at conf, but happy to discuss all this virtually or next time we'll meet!

[cderv]

And nice project on French Data ! I also used to work at RTE so really nice project ! Glad public data are useful ! 😉

[ijlyttle]

Thanks, @cderv - I had forgotten that you had worked at RTE 😳 - I appreciate how open they make things, hoping I can do something useful with it beyond this pipeline!

Too bad you will not be physically at conf - but I will take you up on your offer to discuss virtually! (maybe a little after conf?)

[cscheid]

@ijlyttle There'll be a "posit::conf() lounge", and quarto folks will host a meet-and-greet session on Tuesday, September 19⋅2:20 – 4:00pm CDT. It'd be great to see you!

[ijlyttle]

@cscheid Super, I'll be there!

[DaveParr]

This sounds neat. Hope the talk went well. Do you have any demo code or presentation published on your session? Would love to have a closer look at dvc cml + quarto.