Function to print ggplot2 code for final round of manual adjustment

[kendonB]

Summary

ggrepel seldom gets it right first time. It would be good to be able to have some code to print the ggplot code to copy paste then do some fine adjustment.

[slowkow]

Hi Kendon, thanks for the issue.

Honestly, I typically do the last step (moving labels manually, adjusting font sizes, aligning things) in a program like Adobe Illustrator or Affinity Designer. But I hear you: the more we can do in R, the better.

Since 2016, many users have been trying to extract information from ggplot2 objects to get the ggrepel coordinates: https://github.com/slowkow/ggrepel/issues/24

You might find that one of the replies in #24 has a code snippet that is suitable for you.

The underlying issue is this: the data given to ggplot2 is necessary but insufficient for ggrepel to position labels. We need more information: the dimensions of the plotting area, the dimensions of each label, the dimensions of the data points, etc. The relative sizes of these visual elements is undefined up until the moment that R creates a viewport to render the ggplot2 object to a device. After creating a viewport on a visual device (png, pdf, etc.), that's when the ggrepel coordinates become available.

By the way, that's why ggrepel can automatically reposition labels when we resize the plotting area in X11 or RStudio. It is re-running the repel_boxes2() code again each time the plotting area is resized. So the ggrepel coordinates are dependent on the size of the plotting area.

I have had a hard time trying to implement a built-in function in ggrepel that gives users access to the coordinates. It's been difficult for me to figure out how to do it right, considering the dependency on the viewport.

Thankfully, @alexvpickering forked ggrepel and implemented a function that seems like it might do what you want (I have not tested this code): https://github.com/hms-dbmi/repel/blob/master/R/repel_text.R

@kendonB Could I please ask if you would try the repel_text() function by Alex? (It may need some tweaking.) If you find that this function is useful to you, then maybe we can start a PR to get this merged?

We need to:

• [ ] make a function with a nice user interface like repel_text() by Alex
• [ ] write tests in tests/ to test that it works each time we make changes to ggrepel
• [ ] write at least one example in the documentation to demonstrate how to use it
• [ ] edit NEWS.md to announce your new function
• [ ] edit DESCRIPTION to include your name as a contributor

Pull requests are welcome!

[alexvpickering]

Just a few vague memories from when I worked on the above in case it is helpful:

• repel_text definitely has issues
• it works pretty well for the purpose that I wanted it to, namely, there usually isn't text overlap but I don't care if the text overlaps a point (I am labeling clusters of points in single-cell UMAP plots, not individual points). See https://github.com/hms-dbmi/picker for an example of how I am using it
• it doesn't work so well when you also want the label to not overlap the points. Not sure what I messed up here but even setting ridiculous values like point.padding=30 and point.size=10 doesn't help much and can mess up label overlap
• the too_many_overlaps result is also generally subpar

Happy to experiment a bit if anyone has suggestions on how to improve :)

[slowkow]

Thanks, @alexvpickering for describing your implementation.

To anyone reading this, pull requests are welcome!

Good luck.