stat_summary_bin gives 1 bin more than specified.

[JoFAM]

stat_summary_bin() consistently gives one more bin than specified in the function call

ggplot(data=mtcars, aes(y=mpg,x=wt)) +
  stat_summary_bin(fun.y = "mean", geom = "point",
                   bins = 3)

This shows 4 points. If you change bins to 4, it'll show 5 points and so on.

[yutannihilation]

Maybe the same issue? At least, I feel this needs some better documentation

#1739 (comment)

[JoFAM]

I highly doubt that floating point errors are a reason for a consistent off-by-one error. Especially since this is specific to that function and doesn't occur with hist, cut, ...

[thomasp85]

This is a bug in bin2d_breaks and basically boils down to how we choose binwidth from a range that is closed in both ends

[thomasp85]

we need to extend the range before we calculate the bin width but it is not obvious to me what the best approach is...

Or maybe we should just add a small padding to the last bin so it includes the last data points?

[JoFAM]

@thomasp85 thx for the pointer, didn't have time to dig deeper yet (the stack of function calls to weed through is massive to be honest...). fwiw, cut adds a small padding like this (with nb the number of breaks, 1 more than the number of bins) :

rx <- range(x)
dx <- diff(rx)
breaks <- seq.int(rx[1L], rx[2L], length.out = nb)
breaks[c(1L, nb)] <- c(rx[1L] - dx/1000, rx[2L] + dx/1000)

Judging from the code of bin2d_breaks, adding a similar approach is pretty trivial.

[yutannihilation]

Thanks, I didn't notice this is such a problem... Here's a rendered version of reprex from #1739

library(ggplot2)

x <- seq(0, 1, length = 1e4)
y <- x + rnorm(length(x))
dt <- data.frame(x, y)

# NOT OK
ggplot(dt, aes(x, y)) +
  geom_point(colour = "grey80") +
  stat_summary_bin(fun = mean, bins = 10, geom = "point", colour = "red")

^{Created on 2020-02-21 by the reprex package (v0.3.0)}