Adding limits that contain all data to a histogram generates incorrect missing values warning (due to #2719)

[jmarshallnz]

This seems to have been introduced in 3.1.1 - it does not occur in 3.0.0. I haven't verified completely but I believe it's due to #2719.

Adding scale_x_continuous/xlim to specify limits where all data are contained results in a warning that is incorrect (as no data are missing). The problem is the transformed data (i.e. the bars, xmin/xmax etc) are outside the specified limits, as they are computed using the limits (+/- width). This makes the error quite confusing! The plot is fine - no data are missing as the missing bars contain no data.

library(ggplot2)
#> Warning: package 'ggplot2' was built under R version 3.5.2
set.seed(2)
d <- data.frame(x=runif(1000))
g <- ggplot(d, aes(x=x)) + geom_histogram(bins=10)
g

g + scale_x_continuous(limits=c(-1,2))
#> Warning: Removed 2 rows containing missing values (geom_bar).

Created on 2019-04-24 by the reprex package (v0.2.1)

[thomasp85]

@clauswilke this seem to be an unforeseen consequence of your fix. The only way I can see this being fixed is by having StatBin remove zero-count bins, and I think that will have a lot of unforeseen consequences... alternatively we'd need to subclass GeomBar for geom_histogram() to do some extra work... Thoughts..?

[clauswilke]

I have a number of thoughts:

1. This is broadly related to Should limits be enforced on the calculated geoms? #2887, which we just decided we won't fix at this time.

2. Not sure what the use case is here, but I assume setting limits via coord() would be more appropriate here. Also, it's generally a good idea to set binwidth and center manually for geom_histogram(). We may need to write a tutorial to explain how to get histograms behave nicely.

3. An argument could be made that if only the requested bin number is set (via bins) and explicit limits are set in the scale, then the bins should be chosen such that they fit exactly into the scale limits. This is a fairly easy change to stat_bin() that somebody could implement.

Below I show two examples of how one could plot this example without generating a warning. The first is my preferred solution in this case, because it exactly matches the bins to the known data range. (And it works even if limits are set to 0 to 1).

library(ggplot2)
set.seed(2)
d <- data.frame(x=runif(1000))

ggplot(d, aes(x=x)) + 
  geom_histogram(binwidth = 0.2, center = 0.1) + 
  scale_x_continuous(limits = c(-1, 2))

# this is not entirely appropriate, because the
# bins extend beyond the known data range of [0, 1)
ggplot(d, aes(x=x)) + 
  geom_histogram(bins = 10) + 
  coord_cartesian(xlim = c(-1, 2))

^{Created on 2019-04-24 by the reprex package (v0.2.1)}

[thomasp85]

Ok, as this is covered by #2887 I'm closing

[jmarshallnz]

Thanks! Agreed that #2887 covers this -apologies for missing that one!

I gave the student that spotted this the coord solution before posting.

Agreed that clearly documenting how the limit filtering applies and how coord limiting may be more suited is a decent way to go, as it definitely feels like a change would break things 😀

I wonder if it’s possible to improve the warning? I’ll take a look and see if anything jumps out. Might be able to use Inf for out of scale and add to handle_na?

[lock]

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