Handling of empty strings

[aphalo]

I think it would be best to handle "" as a special case, not drawing segments and empty boxes in this case. It would also help in cases when one wants not to label some observations but still have the observations themselves, e.g., points on the plot, as objects to repel from.

[slowkow]

Thank you for describing the issue and providing images! I like your suggestion!

This merits a detailed example in the documentation to make sure users know how to use this feature.

If we go ahead with this suggestion, then we'll have 3 possibilities for what happens to a data point:

1. The data point has a non-empty string, so it is labeled with a pointer and text.

2. The data point has an empty string, so it is unlabeled, but it also repels other labels. This addresses issue #49.

3. The data point has NA as the string, so it is unlabeled and does not repel other labels. Effectively, it is the same as excluding the data point entirely. Also, R reports a warning about missing values (as it does for geom_text, for example).

   Warning message:
   Removed 1 row containing missing values (geom_text_repel). 
   

[aphalo]

To complement this my new stat can reset the labels to "" based on the density of observations. This would take care of the "static" case, not the "reactive" one, which would need to be handled by text_repel and label_repel on the fly. The images above were produced using the following code:

ggplot(data = d, aes(x, y, label = lab, color = group)) +
geom_point() +
stat_dens2d_labels(geom = "text_repel", 
                              keep.fraction = 0.25,
                              label.fill = "")

and

ggplot(data = d, aes(x, y, label = lab, color = group)) +
geom_point() +
stat_dens2d_labels(geom = "label_repel", 
                             keep.fraction = 0.25,
                             label.fill = "")

In the current version label.fill defaults to NA but I will make "" the default as if the repel geoms are changed as suggested, "" will become the most useful value.

[slowkow]

@aphalo Could I ask you to try 0.6.0? Does this work the way you expect with empty strings?

devtools::install_github("slowkow/ggrepel@0.6.0")

[aphalo]

@slowkow Yes, indeed! Works nicely. Please find attached my examples above rerun with "slowkow/ggrepel@0.6.0".


I also tried larger values for keep.fraction and the plots are fine as long as crowding is not so much as to make satisfactory repositioning impossible.
Many thanks!

[slowkow]

Excellent, and thank you very much for sharing the figures. I'm going to add an example like this to the documentation soon.