R help - Feb 2023 - Palettes {grDevices} - wrong number of colors returned?

Hi,

I would have expected that I get always 3 colors as result which is not 
true:

hcl.colors(3, alpha=c(0, 0.5, 1))     # 3 colors

rainbow(3, alpha=c(0, 0.5, 1))        # 3 colors

heat.colors(3, alpha=c(0, 0.5, 1))    # 3 colors

terrain.colors(3, alpha=c(0, 0.5, 1)) # 6 colors

cm.colors(3, alpha=c(0, 0.5, 1))      # 6 colors

topo.colors(3, alpha=c(0, 0.5, 1))    # 9 colors

R-Version and platform:
R version 4.2.2 Patched (2022-11-10 r83330) -- "Innocent and Trusting"

Copyright (C) 2022 The R Foundation for Statistical Computing

Platform: x86_64-pc-linux-gnu (64-bit)

Bug or feature?

Sigbert

-- 
https://hu.berlin/sk
https://www.stat.de/faqs
https://hu.berlin/mmstat
https://hu.berlin/mmstat-ar

On 23/02/2023 4:36 a.m., Sigbert Klinke wrote:
> Hi,
> 
> I would have expected that I get always 3 colors as result which is not
> true:
> 
> hcl.colors(3, alpha=c(0, 0.5, 1))     # 3 colors
> 
> rainbow(3, alpha=c(0, 0.5, 1))        # 3 colors
> 
> heat.colors(3, alpha=c(0, 0.5, 1))    # 3 colors
> 
> terrain.colors(3, alpha=c(0, 0.5, 1)) # 6 colors
> 
> cm.colors(3, alpha=c(0, 0.5, 1))      # 6 colors
> 
> topo.colors(3, alpha=c(0, 0.5, 1))    # 9 colors
> 
> R-Version and platform:
> R version 4.2.2 Patched (2022-11-10 r83330) -- "Innocent and
Trusting"
> 
> Copyright (C) 2022 The R Foundation for Statistical Computing
> 
> Platform: x86_64-pc-linux-gnu (64-bit)
> 
> Bug or feature?
Looks like a bug to me, they should all be length 3.  The reason it 
happens in terrain.colors (I didn't look at the others) is that it 
breaks the range into two parts, and relies on the number of values of 
the parameters in each part to get the length, but passes the full alpha 
vector in:

terrain.colors <- function (n, alpha, rev = FALSE)
{
     if ((n <- as.integer(n[1L])) > 0) {
	k <- n%/%2
	h <- c(4/12, 2/12, 0/12)
	s <- c(1, 1, 0)
	v <- c(0.65, 0.9, 0.95)
	cols <- c(hsv(h = seq.int(h[1L], h[2L], length.out = k),
                       s = seq.int(s[1L], s[2L], length.out = k),
                       v = seq.int(v[1L], v[2L], length.out = k), alpha 
= alpha),
                   hsv(h = seq.int(h[2L], h[3L], length.out = n - k + 
1)[-1L],
                       s = seq.int(s[2L], s[3L], length.out = n - k + 
1)[-1L],
                       v = seq.int(v[2L], v[3L], length.out = n - k + 
1)[-1L],
                       alpha = alpha))
         if(rev) rev(cols) else cols
     } else character()
}


A bug fix would be to recycle alpha to the right length and only pass in 
portions of it, e.g.

terrain.colors <- function (n, alpha, rev = FALSE)
{
     if ((n <- as.integer(n[1L])) > 0) {
         alpha <- rep_len(alpha, length.out = n)
	k <- n%/%2
	h <- c(4/12, 2/12, 0/12)
	s <- c(1, 1, 0)
	v <- c(0.65, 0.9, 0.95)
	cols <- c(hsv(h = seq.int(h[1L], h[2L], length.out = k),
                       s = seq.int(s[1L], s[2L], length.out = k),
                       v = seq.int(v[1L], v[2L], length.out = k), alpha 
= alpha[seq_len(k)]),
                   hsv(h = seq.int(h[2L], h[3L], length.out = n - k + 
1)[-1L],
                       s = seq.int(s[2L], s[3L], length.out = n - k + 
1)[-1L],
                       v = seq.int(v[2L], v[3L], length.out = n - k + 
1)[-1L],
                       alpha = alpha[seq_len(n - k) + k]))
         if(rev) rev(cols) else cols
     } else character()
}

I'd guess the same sort of approach would fix cm.colors and topo.colors.

Duncan Murdoch

Thanks, Sigbert, good catch.

The reason that this happens is the following: rainbow(), heat.colors(), 
terrain.colors(), cm.colors(), and topo.colors() are all just front-ends 
to calls to hsv() for Hue-Saturation-Value colors. (BTW: This is also the 
main reason why they yield palettes with rather poor perceptual 
properties.)

All of the palette functions just pass on the alpha argument to all 
hsv(..., alpha=alpha) calls. And in each case if the length of alpha and 
the number of colors does not match the shorter element is recycled to the 
length of the longer one.

Thus, if n == length(alpha), as in your example, then those functions that 
can generate all colors with a single call of hsv() do the right thing 
(rainbow and heat.colors). Those, that need two hsv() calls yield 2 * n 
colors (cm.colors and terrain.colors). And in topo.colors() where three 
hsv() calls are used we get 3 * n colors.

My suggested solution would be to do something like alpha <- 
rep_len(alpha, n) early on in the palette functions and then assure that 
the correct elements of alpha are passed to hsv().

The same should be done in those functions that work correctly for n = 
length(alpha) because it will assure that we always get n colors and not 
more. For example, the case with n < length(alpha) is also unexpected:

hcl.colors(2, alpha = c(0, 0.25, 0.5, 0.75, 1))
## [1] "#4B005500" "#FDE33340" "#4B005580"
"#FDE333BF" "#4B0055FF"

Or even worse:

gray.colors(2, alpha = c(0, 0.25, 0.5, 0.75, 1))
## Error in gray(seq.int(from = start^gamma, to = end^gamma, length.out =
n)^(1/gamma),  :
##   attempt to set index 2/2 in SET_STRING_ELT

If someone from R Core thinks that my proposed strategy is a good way 
forward, I'm happy to work on a patch.

Sigbert, if you are looking for a solution: (1) Don't use heat.colors, 
terrain.colors, topo.colors, cm.colors, they are all quite terribel ;-)
(2) If you really want to use them, put on the alpha afterwards, e.g., 
using colorspace::adjust_transparency:

colorspace::adjust_transparency(topo.colors(3), alpha = c(0, 0.5, 1))
## [1] "#4C00FF00" "#00FF4D80" "#FFFF00FF"

hth,
Achim


On Thu, 23 Feb 2023, Sigbert Klinke wrote:
> Hi,
>
> I would have expected that I get always 3 colors as result which is not
true:
>
> hcl.colors(3, alpha=c(0, 0.5, 1))     # 3 colors
>
> rainbow(3, alpha=c(0, 0.5, 1))        # 3 colors
>
> heat.colors(3, alpha=c(0, 0.5, 1))    # 3 colors
>
> terrain.colors(3, alpha=c(0, 0.5, 1)) # 6 colors
>
> cm.colors(3, alpha=c(0, 0.5, 1))      # 6 colors
>
> topo.colors(3, alpha=c(0, 0.5, 1))    # 9 colors
>
> R-Version and platform:
> R version 4.2.2 Patched (2022-11-10 r83330) -- "Innocent and
Trusting"
>
> Copyright (C) 2022 The R Foundation for Statistical Computing
>
> Platform: x86_64-pc-linux-gnu (64-bit)
>
> Bug or feature?
>
> Sigbert
>
> -- 
> https://hu.berlin/sk
> https://www.stat.de/faqs
> https://hu.berlin/mmstat
> https://hu.berlin/mmstat-ar
>
> ______________________________________________
> R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>