stat_density() ignores weight

[eliocamp]

While both geom_density() and stat_density() compute weighted densities correctly, the later results in an spurious warning (the first of the bunch).

library(ggplot2)

df <- data.frame(x = runif(100))

ggplot(df, aes(x)) +
   stat_density(aes(weight = abs(x - 0.5))) +
   geom_density(aes(weight = abs(x - 0.5)), color = "red")
#> Warning: Ignoring unknown aesthetics: weight
#> Warning in density.default(x, weights = w, bw = bw, adjust = adjust, kernel
#> = kernel, : sum(weights) != 1 -- will not get true density

#> Warning in density.default(x, weights = w, bw = bw, adjust = adjust, kernel
#> = kernel, : sum(weights) != 1 -- will not get true density

Created on 2018-09-19 by the reprex package (v0.2.0).

(I'm sorry if the multiple similar issues are not cool; you can tell that I'm working with weighted densities in my work and finding these small issues 😆 )

[clauswilke]

It's better to file separate issues for separate items, thanks!

[paleolimbot]

It looks like GeomDensity sets a default aesthetic for weight, which silences the warning (even though weight isn't in the data that gets passed to GeomDensity.

ggplot2/R/geom-density.r

Lines 75 to 80 in 1f6f0cb

	GeomDensity <- ggproto("GeomDensity", GeomArea,
	default_aes = defaults(
	aes(fill = NA, weight = 1, colour = "black", alpha = NA),
	GeomArea$default_aes
	)
	)

I very briefly tried adding weight = 1 to the default aesthetics of StatDensity, which does silence the warning (and doesn't cause any tests to fail).

ggplot2/R/stat-density.r

Lines 65 to 67 in 1f6f0cb

	StatDensity <- ggproto("StatDensity", Stat,
	required_aes = "x",
	default_aes = aes(y = stat(density), fill = NA),

[hadley]

Two problems:

• We need to do whatever is necessary to tell stat_density() that weight is a valid aesthetic.

• In compute_density(), I think we should always do w <- w / sum(w) to silence the warning from density().

[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/