impossiblewizardry | (no subject)

When I published an R function, I mean made an R library which was just this one function and put it on a repository, I made sure it was accurate no matter what input you put in.

I didn’t need this level of accuracy. I was only using the function to get outputs that were around .01, or maybe .001. I didn’t need accuracy when the outputs were smaller than that. This is part of why publishing the function took a lot of effort. Writing it for my own use was comparatively easy. The extra stuff I had to do before I published it took, I think, longer than writing the function for my own use in the first place.

And it’s online for I think less than a month, when someone is e-mailing me asking if I can add an option to give answers on a log scale. So I guess this person is getting answers from my function that end up getting rounded to 0 due to limited floating point precision, and they want to know how low it really is.

So even though I told myself I was making it usable for everyone, I actually underestimated the level of accuracy that someone out there would need. I still don’t understand what for. Based on her publications, I’d guess she isn’t using it for data analysis, but rather as a component in some kind of.. physics simulation or something? I don’t get it, and didn’t ask for details, just told her that the algorithm I’m currently using couldn’t support it, here’s how she might do it on her own, and I’l let her know if I make any changes.

Anyway... the point is when you do math stuff you don’t know how people are going to use it. And when you do applied math stuff you really want people to be able to treat your tool as a black box, and be able to trust the answers they get without a deep understanding of how they’re being produced, and figuring out what inputs they’ll get accurate answers, and what inputs they won’t.

Which means, from a mathematical perspective, either you need exact answers, or you need bounds on the answer.

So when I read Zeilberger write stuff like "But let’s be pragmatic and forget about our purity and obsession with ‘exact’ answers.” (source) He has no idea what’s pragmatic.