size in geom_point using variable with all but 1 as NAs

[iagomosqueira]

When passing a variable in the data.frame to the size argument in geom_point, if the column to be all but one as NA, all values get plotted equally, i.e. NAs are not dropped. If two non-NA values exist, the plot looks as expected.

Wanted to know if it is an unavoidable consequence of how size is calculated, or really a bug. It the later I could try finding/fixing it.

  df <- data.frame(year=rep(10:14, each=3), age=1:3, data=runif(15))

  # points sized as expected
  ggplot(df, aes(year, age)) + geom_point(aes(size=data))

  # convert all values but 2 into NAs
  df$data[1:13] <- NA
  #2 points sized and shown
  ggplot(df, aes(year, age)) + geom_point(aes(size=data), na.rm=TRUE)

  # convert all but one value into NA
  df$data[1:14] <- NA

  # plot shows equal-sized dots for all datapoints
  ggplot(df, aes(year, age)) + geom_point(aes(size=data), na.rm=TRUE)

I tried with different names for the data column, but it made no difference. Tested in both ggplot2 1.0.0 from CRAN and 1.0.0.99 from github.

Thanks

[hadley]

That is a strange one. I'll take a look. Even more minimal reprex:

df <- data.frame(x = rep(1:2, each = 2), y = rep(1:2, 2), z = c(1, NA, NA, NA))
ggplot(df, aes(x, y, size = z)) + geom_point()

[hadley]

Ah, this is a bug in scales::rescale():

function(x, to = c(0, 1), from = range(x, na.rm = TRUE)) {
  if (zero_range(from) || zero_range(to)) return(rep(mean(to), length(x)))

  (x - from[1]) / diff(from) * diff(to) + to[1]
}

If from has zero_range(), it needs to preserve missing values in x.

[hadley]

Fixed by r-lib/scales@b481ee2