Default color palette

[thibautjombart]

The 3 requirements of the new color palette would be:

• look nice to as many humans as possible
• be colorblind friendly
• correspond to categorical variables

Quite a bit of thinking on these has been done by the viridis package:
https://cran.r-project.org/web/packages/viridis/vignettes/intro-to-viridis.html

As well, these very good resources:

• https://personal.sron.nl/~pault/#sec:qualitative
• https://www.osapublishing.org/oe/abstract.cfm?uri=oe-21-8-9862

I suggest we use this issue to propose palettes. Ideally put them to a vote at some point.

[TimTaylor]

I'm colourblind (red/green) so hopefully help out with the testing of palettes.

With the release of R 4.0 (see https://developer.r-project.org/Blog/public/2019/11/21/a-new-palette-for-r/index.html). Sometimes I find the most colourblind-friendly palettes are not necessarily the nicest looking (this is subjective of course). One additional thing to think about is palette ordering. Sometimes you can get away with a less friendly palette if you are only using a small subset of them.

[thibautjombart]

Not that I'm nessarily advocating for it, but here's one, derived from 'paired' in RColorBrewer, which natively allows up to 12 levels; not color-blind friendly I fear:

## trying colors
### paired tweaked
incidence_pal <- function(n) {
  col <- c("#1F78B4", "#33A02C", "#E31A1C", "#FF7F00", "#6A3D9A", "#B15928", 
           "#A6CEE3", "#B2DF8A", "#FB9A99", "#FDBF6F", "#CAB2D6", "#FFFF99"
           )
  if (n < length(col)) {
    return(col[seq_len(n)])
  }

  colorRampPalette(col)(n)
}


barplot(1:12, col = incidence_pal(12), main = "palette: sorted `paired`")

## 3 colors
barplot(matrix(sample(1:10, 1000, replace = TRUE), nrow = 3, ncol = 60),
        col = incidence_pal(3), border = "transparent")
#> Warning in matrix(sample(1:10, 1000, replace = TRUE), nrow = 3, ncol = 60): data
#> length [1000] is not a sub-multiple or multiple of the number of rows [3]

## 8 colors
barplot(matrix(sample(1:10, 1000, replace = TRUE), nrow = 8, ncol = 60),
        col = incidence_pal(8), border = "transparent")
#> Warning in matrix(sample(1:10, 1000, replace = TRUE), nrow = 8, ncol = 60): data
#> length [1000] is not a sub-multiple or multiple of the number of columns [60]

## 12 colors
barplot(matrix(sample(1:10, 1000, replace = TRUE), nrow = 12, ncol = 60),
        col = incidence_pal(12), border = "transparent")
#> Warning in matrix(sample(1:10, 1000, replace = TRUE), nrow = 12, ncol = 60):
#> data length [1000] is not a sub-multiple or multiple of the number of rows [12]

^{Created on 2020-07-03 by the reprex package (v0.3.0)}

[TimTaylor]

From left to right, drop colours 1, 2, 3 and 7 and it works for me

incidence_pal <- function(n) {
  col <- c("#FF7F00", "#6A3D9A", "#B15928", "#B2DF8A", "#FB9A99", "#FDBF6F", "#CAB2D6", "#FFFF99"
  )
  if (n < length(col)) {
    return(col[seq_len(n)])
  }
  
  colorRampPalette(col)(n)
}


barplot(1:8, col = incidence_pal(8), main = "palette: reduced sorted `paired`")

^{Created on 2020-07-03 by the reprex package (v0.3.0)}

[thibautjombart]

This one is derived from the colorbling-friendly categorical variable palettes at: https://personal.sron.nl/~pault/#sec:qualitative
Two versions:

• vibrant: my favourite, less categories
• muted: nice too, more categories

I think I also like the colors better, but how does it work for you?

## my preferred, uses https://personal.sron.nl/~pault/#sec:qualitative
col_vibrant <- c(
  "#0077BB",
  "#33BBEE",
  "#009988",
  "#EE7733",
  "#CC3311",
  "#EE3377",
  "#BBBBBB"
  )

col_muted <- c(
  "#332288",
  "#88CCEE",
  "#44AA99",
  "#117733",
  "#999933",
  "#DDCC77",
  "#CC6677",
  "#882255",
  "#AA4499",
  "#BBBBBB"
  )


make_palette <- function(x) {
  function(n) {
    if (length(x) > n) {
      x[seq_len(n)]
    } else {
      colorRampPalette(x)(n)
    }
  }
}


#### vibrant palette
pal_vibrant <- make_palette(col_vibrant)
barplot(1:7, col = pal_vibrant(7), main = "palette: vibrant")

barplot(matrix(sample(1:10, 1000, replace = TRUE), nrow = 3, ncol = 60),
        col = pal_vibrant(3))
#> Warning in matrix(sample(1:10, 1000, replace = TRUE), nrow = 3, ncol = 60): data
#> length [1000] is not a sub-multiple or multiple of the number of rows [3]

barplot(matrix(sample(1:10, 1000, replace = TRUE), nrow = 8, ncol = 60),
        col = pal_vibrant(8))
#> Warning in matrix(sample(1:10, 1000, replace = TRUE), nrow = 8, ncol = 60): data
#> length [1000] is not a sub-multiple or multiple of the number of columns [60]

#### muted palette
pal_muted <- make_palette(col_muted)
barplot(1:7, col = pal_muted(7), main = "palette: muted")

barplot(matrix(sample(1:10, 1000, replace = TRUE), nrow = 3, ncol = 60),
        col = pal_muted(3))
#> Warning in matrix(sample(1:10, 1000, replace = TRUE), nrow = 3, ncol = 60): data
#> length [1000] is not a sub-multiple or multiple of the number of rows [3]

barplot(matrix(sample(1:10, 1000, replace = TRUE), nrow = 10, ncol = 60),
        col = pal_muted(10))
#> Warning in matrix(sample(1:10, 1000, replace = TRUE), nrow = 10, ncol = 60):
#> data length [1000] is not a sub-multiple or multiple of the number of columns
#> [60]

^{Created on 2020-07-03 by the reprex package (v0.3.0)}

[TimTaylor]

Vibrant is best for me. Which colour would you use as the default for facets and ungrouped plots?

[thibautjombart]

Awesome. Maybe the first of the palette as default color for single plots / no groups? This said, feel free to reorder the colors of the palette if you fancy.

[TimTaylor]

Now implemented. What do you think about the default for the column borders (white or NA)

[thibautjombart]

Morning! I was trying to test both to compare and found this issue. I think it may come from the borders (bars are so thin we see only the border)? If that is the case, that could be an element to consider in the default. My thinking would be:

• transparent border: safer for thin bars (if that's the issue)
• grey / black border: may be better contrast between colors (not even sure if that's true)

[thibautjombart]

On an unerlated point: maybe it would make sense to plot the missing (NA) group as grey?

[TimTaylor]

Are we happy to close this now?