Vignette on transforming distributions

[harriet-mason]

I actually didn’t realise you could do operations like 5 + dist_normal until I read your talk slides. The talk slides are a bit hidden away and hard to find. If you don’t have plans to write a vignette, I could just collect the information from your talks and make one and make a pull request for it (not right now... but eventually). I think a description of the philosophy of the package would be beneficial to new users who are kind of alien to this idea.

[mitchelloharawild]

Thanks for volunteering to contribute a vignette 🎉

Vignettes are certainly needed for the package, preferably they are concise and focused on specific aspects of the package (e.g. transformed distributions which would cover Ops, Math, and dist_transformed() functionality). Feel free to adapt content from these slides: https://slides.mitchelloharawild.com/user2024

A package philosophy vignette is probably best written by myself (or in a paper), it would also encompass argument recycling based on I/O described in #52 (comment) (already noted in #107)

[harriet-mason]

@mitchelloharawild yeah, I agree, sorry I was unclear. I meant that I am happy to write a vignette on the transformation because I can just copy over the slides, but separately, there needs to be a vignette on the package philosophy. While I definitely have faith you will finish the paper and certainly don't think you will over-leverage yourself and leave the explanation in purgatory forever, maybe writing a brief, high-level explanation in the interim would be useful?

I mentioned this because my main difficulty in learning to use the package was anticipating how the functions were supposed to work. Personally, I am not as software-brained as all you guys having a powwow in the GitHub issues, so discussions about vector recycling and how functions should be applied go over my head. I've got a real good "package go brrrr" thing going on that I don't want to mess with.

My understanding of how the package should work is mostly based on the package I would like to have, where the "units" for the column are just random variables, but I have no idea how well this aligns with the Mitch/Matt philosophy. That is why I have no idea if the factor(dist) = dist_transformed(dist, factor, as.numeric) is actually how the package should work. The same thing goes for other functionality I would like. For example, I think that dist_normal(0,1)>=0 should return a dist_bernouli(0.5). My logic is that a numeric>0 results in a logical, and a random logical is a Bernoulli random variable; however, I can also understand someone else thinking it should return a dist_truncated(dist_normal(0,1), lower = 0) because of how they think distributional should work in that scenario. As far as I know, what you think it should do could be something wildly different. If people understand the way the package should behave, it makes it easier to know when it is misbehaving (or when that functionality hasn't been built out yet) and make an issue/pull request to fix it and improve the package.

Either way, I am happy to add a vignette with the transformation info from the slides, but I do think a higher-level explanation of the philosophy will make pull requests/issues easier and reduce your workload in the long run.

[mitchelloharawild]

Let's try to keep issues specific to a single topic, I'll open a new issue for a vignette / paper on design philosophy.