search

tidymodels / spatialsample

Create and summarize spatial resampling objects 🗺
https://spatialsample.tidymodels.org
Other
71 stars 5 forks source link

Update getting started vignette #69

Closed mikemahoney218 closed 2 years ago

mikemahoney218 commented 2 years ago

This PR closes #27 by refreshing the getting started vignette to incorporate the newest changes to spatialsample.

juliasilge commented 2 years ago

Is the spatially-buffered CV GIF supposed to look like this?

buffer

I'm having a hard time interpreting this one, even with the text. If this is right, can we say just a little bit more?

mikemahoney218 commented 2 years ago

Gah!

So, I spent a lot of time fussing with the right buffer to use to induce spatial extrapolation, and basically failed. I tried using a buffered LOO approach, but there's a handful of houses the model much worse on, which screws up the color scale for all the plots at the end. So I moved back to not changing v and just running a larger buffer, attempting to get some amount of spatial extrapolation, but didn't think about what that would do to the GIF. (It also didn't really get any extrapolation in the final result).

Using a smaller buffer (in this case, 50) makes a different GIF -- is this more useful? Ames is so dense/overplotted that I'm not entirely sure this is easy to understand, either:

ArGBAAOw==

juliasilge commented 2 years ago

RE: the buffering GIF, I really felt like I understood the buffering after the other vignette but now the GIF makes me suspect I don't know how this is supposed to work. 🙈 I do think making the buffer much smaller helps.

Would it be accurate to say something like:

Some of these home are very close together so it may be difficult to see each individual home because of overplotting. There is an assessment set point everywhere you see a cluster of buffered points. Points inside of this buffer are included in neither the analysis ...

It's not... perfect still. Do you think the value of showing spatial_buffer_vfold_cv() in the main vignette makes it OK to show something that's not super duper clear? Or is this a situation where you probably wouldn't use spatial_buffer_vfold_cv() so it's confusing to readers?

mikemahoney218 commented 2 years ago

I think it is dissatisfying to show 3 out of 4 methods in the introductory vignette, but worse to try and introduce spatial_buffer_vfold_cv() here in the way we're doing it. I'd be content with letting the buffering vignette be the entire introduction to the concept of buffering, rather than trying to halfway explain it here.

mikemahoney218 commented 2 years ago

Though as a second thought, we are leaving the door open to adding more methods in the future, and it's probably good to not establish the precedent that all methods must be in the getting started vignette. I think it's fine to not include it here.

github-actions[bot] commented 2 years ago

This pull request has been automatically locked. If you believe you have found a related problem, please file a new issue (with a reprex: https://reprex.tidyverse.org) and link to this issue.