Request: support providing shapes as names rather than integers

[daattali]

Currently if I want to use ggplot and make the points a certain shape, I need to say "shape=5" for example. Magic numbers like that are inconvenient: the author needs to figure iut what number means what shape, and the person reading the code has no idea what shape that'll be.

It'd be awesome to be able to say "geom_point (shape='diamond')

@thomasp85

[daattali]

Suggestion (I don't have any strong opinion about the naming, just putting this out there so that others can improve on it or implement it):

0: square-open
1: circle-open
2: triangle-up-open
3: cross
4: x-mark (I didn't want to use simply "x" because currently using "x" or any other letter actually uses that letter)
5: diamond-open
6: triangle-down-open
7: square-x
8: asterisk
9: diamond-x
10: circle-cross
11: star
12: square-cross
13: circle-x
14: square-triangle
15: square
16: circle
17: triangle-up
18: diamond
19: circle (I can't figure out the difference between this and 16)
20: circle-small
21: circle-fill
22: square-fill
23: diamond-fill
24: triangle-up-fill
25: triangle-down-fill

[has2k1]

@daattali, If I recall correctly, 16 & 19 can vary in appearance on some OSes depending on the underlying devices. e.g one of them may be aliased and the other not aliased.

[smouksassi]

from the documentation ggplot shapes are like pch base graphics parameter and in the pch help we have:
The difference between pch = 16 and pch = 19 is that the latter uses a border and so is perceptibly larger when lwd is large relative to cex.

[daattali]

Thanks Samer :)

[hturner]

I have a few suggestions for improvements:

• As the names will be specified as character strings, you may as well allow spaces for readability - otherwise perhaps use snake_case for general consistency with ggplot2 conventions.
• Aim for consistency with the naming in ?pch, i.e. describe fill first, use same names for shapes. However, I like your suggestion to have some defaults to avoid long names - as well the default being a solid shape, I suggest assuming triangles are point up unless otherwise stated.
• Use "plus" and "cross" vs "cross" and "x-mark": "plus" is unambiguous, "cross" can imply different shapes, but is often used for an x mark and works better for combinations of shapes.
• If we think of 14 as a square with a chevron on top, we can have the convention that the closed shape always comes first (incidentally the orientation of the triangle comes out differently when I plot it from how it is in ?pch - different OS? different R version? Anyway, chevron works both ways).

Putting all together:

0: open square
1: open circle
2: open triangle
3: plus
4: cross
5: open diamond
6: open triangle down [or point down for consistency with ?pch]
7: square cross
8: asterisk
9: diamond plus
10: circle plus
11: star
12: square plus
13: circle cross
14: square chevron
15: square
16: small circle [as this is the one without the border, smaller than 1 and 19]
17: triangle
18: diamond
19: circle
20: bullet
21: filled circle
22: filled square
23: filled diamond
24: filled triangle
25: filled triangle down [or point down]

[hadley]

This would be a pull request if someone else wanted to implement it. You'd need:

• A function that translated names to numbers (probably using match.arg())
• To call that function in geom_point()

[hturner]

This could be a "good first issue" - now that GitHub is promoting this label to help potential first-time contributors discover issues, it would be good to see this being used in the R community. An accessible ggplot2 issue would be a good way to promote this.

[hadley]

Agreed - I don't have the resources right now to do that systematically, but I will in the near future.

[daniel-barnett]

I've implemented this so far with the strings from @hturner in snake case. Does anybody else have any suggestions/opinions on some of the strings such as "point down" vs simply "down", etc? "filled_triangle_point_down" feels like it's getting a bit lengthy to my eyes, so I've went with "down" for now.

[ptoche]

symbols 21, 22, 23, 24, 25 are special in that they work with both fill and color, whereas the other symbols will ignore fill. Is there a case for making them a more convenient group of symbols to call? e.g. naming them simply "circle", "square", etc. without prefixing with "filled"...

[daniel-barnett]

It depends how we want to treat the relation between the default shape set (which do not fill) and what the strings correspond to.

[daniel-barnett]

To make people's lives easier, here goes the proposed correspondence between strings and shape numbers:

library(ggplot2)

.pch_table <- c("0" =  "open_square",
                "1" =  "open_circle",
                "2" =  "open_triangle",
                "3" =  "plus",
                "4" =  "cross",
                "5" =  "open_diamond",
                "6" =  "open_triangle_down",
                "7" =  "square_cross",
                "8" =  "asterisk",
                "9" =  "diamond_plus",
                "10" = "circle_plus",
                "11" = "star",
                "12" = "square_plus",
                "13" = "circle_cross",
                "14" = "square_triangle",
                "15" = "square",
                "16" = "small_circle",
                "17" = "triangle",
                "18" = "diamond",
                "19" = "circle",
                "20" = "bullet",
                "21" = "filled_circle",
                "22" = "filled_square",
                "23" = "filled_diamond",
                "24" = "filled_triangle",
                "25" = "filled_triangle_down")

df_shapes <- data.frame(shape = 0:25, shape_name = factor(paste0(0:25, " ('", .pch_table, "')")))

ggplot(df_shapes, aes(0, 0, shape = shape)) +
  geom_point(aes(shape = shape), size = 5, fill = 'red', stroke = 2) +
  scale_shape_identity() +
  facet_wrap(~reorder(shape_name, shape)) +
  theme_void()

[ptoche]

Nice. There is room for disagreement about the relative merits of filled_circle and circle_filled or filled.circle or circle.filled. I was under the impression that x_y was more for functions, e.g. geom_boxplot while x.y was more for function arguments, e.g. outlier.shape. I could be wrong. Also there are situations where, say if you import a spreadsheet, a circle filled gets converted to circle.filled so possibly some benefits from having dots. I have no strong opinion. Personally I rarely use shapes outside of 21-26 and have them memorized.

[hturner]

I don't have a particular preference for filled_circle vs filled.circle - I'm happy for a regular contributor to the tidyverse to say which is more consistent with their style.

The choice to use "filled" for symbols 21-25 and to use this as prefix rather than a suffix was based on the use in ?pch, to quote:

The following R plotting symbols are can be obtained with pch = 19:25: those with 21:25 can
be colored and filled with different colors: col gives the border color and bg the background
color (which is "grey" in the figure)
  - pch = 19: solid circle,
  - pch = 20: bullet (smaller solid circle, 2/3 the size of 19),
  - pch = 21: filled circle,
  - pch = 22: filled square,
  - pch = 23: filled diamond,
  - pch = 24: filled triangle point-up,
  - pch = 25: filled triangle point down.

[hadley]

Given that these are strings, I'd prefer "filled circle", and I think the modifiers would be better as suffixes (so when you sort alphabetically so you see related shapes close together), so that would lead to "circle filled".

[daniel-barnett]

Thanks for the input, everyone.

Here goes the new names. I went with triangle down filled (opposed to triangle filled down) as I thought it's a bit more logical in that triangle and triangle down are (slightly) different shapes.

library(ggplot2)

pch_table <- c("0" = "square open",
               "1" = "circle open",
               "2" = "triangle open",
               "3" = "plus",
               "4" = "cross",
               "5" = "diamond open",
               "6" = "triangle down open",
               "7" = "square cross",
               "8" = "asterisk",
               "9" = "diamond plus",
               "10" = "circle plus",
               "11" = "star",
               "12" = "square plus",
               "13" = "circle cross",
               "14" = "square triangle",
               "15" = "square",
               "16" = "circle small",
               "17" = "triangle",
               "18" = "diamond",
               "19" = "circle",
               "20" = "bullet",
               "21" = "circle filled",
               "22" = "square filled",
               "23" = "diamond filled",
               "24" = "triangle filled",
               "25" = "triangle down filled")

df_shapes <- data.frame(shape = 0:25, shape_name = factor(paste0(0:25, " ('", pch_table, "')")))
df_shapes <- df_shapes[order(pch_table),]

ggplot(df_shapes, aes(0, 0, shape = shape)) +
  geom_point(aes(shape = shape), size = 5, fill = 'red', stroke = 2) +
  scale_shape_identity() +
  facet_wrap(~reorder(shape_name, shape)) +
  theme_void()

sort(pch_table)

                     8                     20                     19 
            "asterisk"               "bullet"               "circle" 
                    13                     21                      1 
        "circle cross"        "circle filled"          "circle open" 
                    10                     16                      4 
         "circle plus"         "circle small"                "cross" 
                    18                     23                      5 
             "diamond"       "diamond filled"         "diamond open" 
                     9                      3                     15 
        "diamond plus"                 "plus"               "square" 
                     7                     22                      0 
        "square cross"        "square filled"          "square open" 
                    12                     14                     11 
         "square plus"      "square triangle"                 "star" 
                    17                     25                      6 
            "triangle" "triangle down filled"   "triangle down open" 
                    24                      2 
     "triangle filled"        "triangle open" 

[hadley]

Can you please redo that plot alphabetically ordering the shapes? I think your code will be simpler if you flip the names and values.

[ptoche]

like this?

pch_table2 <- sort(setNames(names(pch_table), unname(pch_table)))

     square open          circle open 
             "0"                  "1" 
     circle plus                 star 
            "10"                 "11" 
     square plus         circle cross 
            "12"                 "13" 
 square triangle               square 
            "14"                 "15" 
    circle small             triangle 
            "16"                 "17" 
         diamond               circle 
            "18"                 "19" 
   triangle open               bullet 
             "2"                 "20" 
   circle filled        square filled 
            "21"                 "22" 
  diamond filled      triangle filled 
            "23"                 "24" 
triangle down filled                 plus 
            "25"                  "3" 
           cross         diamond open 
             "4"                  "5" 
  triangle down open         square cross 
             "6"                  "7" 
        asterisk         diamond plus 
             "8"                  "9" 

[hadley]

I meant just redraw the plot alphabetically.

[ptoche]

I'm getting warnings... but the output looks ok:

df_shapes <- data.frame(shape = 0:25, 
                        shape_name = pch_table, stringsAsFactors = FALSE)
df_shapes <- df_shapes[order(df_shapes$shape_name),]
df_shapes$shape_name <- factor(df_shapes$shape_name)
ggplot(df_shapes, aes(0, 0, shape = shape)) +
    geom_point(aes(shape = shape), size = 5, fill = 'red', stroke = 2) +
    scale_shape_identity() +
    facet_wrap(~ shape_name) +
    theme_void()

[hadley]

That's great, thanks!

I wonder if star should be triangle up down? And maybe it would be useful to have manually specified rows and columns? Then you could (e.g.) display the cross and plus variations in the matching columns. This would be useful to help people understand how the shapes are related to one another.

[ptoche]

Together with "triangle down up" <- "triangle up down" ?

and

"diamond plus" <- "plus diamond"
"square cross" <- "cross square"
"square plus" <- "plus square"
"square triangle" <- "triangle square"

[ptoche]

do you mean a grid like this one?

or basic shapes along both directions for a triangular matrix effect?

[hturner]

Having the modifiers as suffixes makes more sense to me when the shapes are ordered alphabetically. But this leads me to suggest a couple of changes:

• "circle tiny" instead of "bullet"?
• "nabla" instead of "triangle down"? (https://en.wikipedia.org/wiki/Nabla_symbol). This is a lot shorter to type and if you do sort the shapes alphabetically, the nablas will be separate from the triangles. Another way to separate the two types of triangle when sorting alphabetically would be to use "triangle rotated" but this is a lot longer, especially with the filled/open suffix.

I still like "star as it is short and describes the shape quite well. But "nabla triangle"/"triangle nabla" also works.

I think Hadley's last comment about the plot was to arrange so that the row represented the base shape and the column represented the modifier, e.g. the layout would be (shortened example)

     [,1]       [,2]           [,3]              [,4]           
[1,] "plus"     ""             ""                ""             
[2,] "square"   "square cross" "square filled"   "square open"  
[3,] "triangle" ""             "triangle filled" "triangle open"

In which case alphabetical ordering is less important, but having a one-word name for the base shape is quite useful.

[ptoche]

"circle tiny" is more consistent with the naming convention shape+qualifier. By this reasoning, "triangle down" works better than "nabla".

[hadley]

@ptoche I mean a grid where you've carefully arranged the cells by hand so that related symbols appear next to each other. For example, in the previous plot, it would be nice if "triangle filled" and "triangle open" appeared on the far right to align with the filled and open shapes. Similarly, it would be nice if the all the unmodified shapes were arranged in one column. This would make it easier to see the underlying pattern in the shapes.

[baptiste]

@hadley the problem is that many shapes belong to more than one family (e.g. both triangle and square),

[hadley]

@baptiste there's no reason for a shape to appear in only one place 😉

[baptiste]

There we go again with quantum Venn diagrams ;)

I think no matter how one puts it, these shapes are an odd bunch. Years ago I thought it would be nice to get a better set included in the low-level graphics primitives (more consistent, in terms of sizes, attributes, combinations, redundancy, etc.), but that seems unlikely nowadays; instead I hope that something like svg will soon become the only graphics format worth discussing, and all these 'device' quirks will be left behind.

[clauswilke]

Can this issue be closed? It seems to me that this has been addressed with #2338. The following works for me.

library(ggplot2)
mtcars$am2 <- ifelse(mtcars$am, "Manual", "Automatic")
ggplot(mtcars) + geom_point(aes(hp, cyl, shape = am2)) + 
  scale_shape_manual(values = c("Manual" = "triangle open", "Automatic" = "plus")) 

Created on 2018-05-11 by the reprex package (v0.2.0).

[hadley]

Oops yes

[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/