Elektrine lite

← Feed

@RossGayler@aus.social

2026-10-24 06:04 UTC

@vincentab@fosstodon.org Hi Vincent - where did you end up with this question? FWIW when I was doing large model development projects (years ago) I ended up with multiple, long R-markdown notebooks. Human readable = yes, machine readable = no (-ish). Questions and decisions were highlighted in numbered text boxes. I also kept a data dictionary (as an R data frame) that evolved programmatically over the course of the analyses. New variables were added to the data dictionary as a they were generated in the analyses and the variables were tagged with analysis-relevant properties (so I could, for example, programmatically select variables to use as predictors in a model). The analyses were reproducible by re-running the notebooks. This was pre [targets] which I would probably use now. At the time I used [workflowr]. I suspect this was less structured than you were looking for. The issue is that it mostly realied on me trying really hard to be systematic and reflect all the decisions in the free text.

Replies (1)

  • @vincentab@fosstodon.org 2026-10-24 12:00

    @RossGayler@aus.social That's interesting! A bit ago I worked on something called "Data Analysis Decision Records", closely following the ADR that software engineers use. I even coded up an app an everything. But I tried to post it a few times and it never seem to quite catch on. It's too bad, because I think the idea has a lot of potential. Maybe I should invest in community building a bit. I'd be curious to know what you think! https://vincentarelbundock.github.io/dadr/

    Open ##5300352