gnuplot: plot points with color based values in one string column and show string in legend

Question

I would like to plot results of classification and mark true classes. So, basically what I need is to assign a color for each point base on value in a string column.

dataset looks like this:

5.1 3.5 1.4 0.2 Iris-setosa

I ended up with script following solution (thanks to the answer in here: How to make points one color when a third column equals zero, and another color otherwise, in Gnuplot?)

set palette model RGB defined (0 "red",1 "blue", 2 "green")
plot 'iris.data' using 1:2:5 notitle with points pt 2 palette

in the original dataset I replaced string labels with numbers, because I don't know how to work with strings in gnuplot. Is there a way how to map string to colors?

Currently the output looks like this:

However I don't like the gradient palette because it doesn't make sense in this case. I would prefer normal legend with a single color and name of the class. Any idea how to do that?

Answer 1

As colors are hex numbers, then we can take some starting color and some increasing step, for example I took in bash

FIRST_COLOR=0xB97900
ADDING_COLOR=0x1F541E

Then, in for cycle, for lines I used hex numbers addition

for ((i = 1 ;  i <=  $some_number ; i++ ));
do
 FIRST_COLOR=$(( $FIRST_COLOR + $ADDING_COLOR ))
 x=`printf "0x%X\n" $FIRST_COLOR`
 echo "set style line $i lc rgb '$x' lt 1 lw 1 pt $i ps 1" > some_file
done

so, we can have in some_file as many colors as want, though if the initial number and increment are poorly selected, then the colors are difficult to distinguish, although a lot depends on the quality of the screen.

Same can be done, of course, for points.

Hope, helps somebody.

Answer 2

Mapping strings to colors or to palette indices can be done natively in gnuplot, the key is to use stringcolumn() and a user-defined mapping function.

Here is an example using a palette:

#!/usr/bin/gnuplot -persist

# define a palette with an exact number of colors
set palette maxcolors 3
set palette model RGB defined ( \
  0 "red", \
  1 "blue", \
  2 "green")

# Define palette labels, the range matches the number of colors defined above
set cbrange [0:3]
set cbtics offset 0,+4 ( \
  'color1' 0, \
  'color2' 1, \
  'color3' 2, \
  '' 3)

# define a function to map strings to palette indices
map_color(string) = ( \
  string eq 'color1' ? 0 : \
  string eq 'color2' ? 1 : \
  string eq 'color3' ? 2 : \
  3)

plot '-' using 1:2:(map_color(stringcolumn(3))) notitle with points pt 2 palette
5.4452 4.6816 color1
1.2079 9.4082 color2
7.4732 6.5507 color2
2.3329 8.2996 color2
3.4535 2.1937 color3
1.7909 2.5173 color3
2.5383 7.9700 color1
EOF

A similar approach can be taken to map to colors directly:

#!/usr/bin/gnuplot -persist

# define a function to map strings to palette indices
map_color(string) = ( \
  string eq 'color1' ? 0xff0000 : \
  string eq 'color2' ? 0x0000ff : \
  string eq 'color3' ? 0x00ff00 : \
  0x000000)

plot '-' using 1:2:(map_color(stringcolumn(3))) notitle with points pt 2 lc rgbcolor variable
5.4452 4.6816 color1
1.2079 9.4082 color2
7.4732 6.5507 color2
2.3329 8.2996 color2
3.4535 2.1937 color3
1.7909 2.5173 color3
2.5383 7.9700 color1
EOF

Answer 3

A color palette can be used to get any point color (for individual points)

plot file using 1:2:3 with points palette

Now set up a palette that gives you the desired color scale. You can set the palette to compute the color for example using the HSV color model or have a look at the the gnuplot demos website.

Just for completeness, have a look at the last example on gnuplot pm3d colors demo page:

theta(x) = x<0 ? 0 : 1
r(x) = 4*x*(1-theta(x-0.25))
g(x) = 0.5*theta(x-0.25)*(1-theta(x-0.5))
b(x) = x
set palette model RGB functions r(gray),g(gray),b(gray)
set title "set palette model RGB functions\n4*x*(1-theta(x-0.25)), 0.5*theta(x-0.25)*(1-theta(x-0.5)), x"
splot f(x)enter code here

Answer 4

A way how you could do that is by using awk.

Using a data file Data.csv:

5.4452 4.6816 blue
1.2079 9.4082 red
7.4732 6.5507 red
2.3329 8.2996 red
3.4535 2.1937 green
1.7909 2.5173 green
2.5383 7.9700 blue

and this script:

set pointsize 3
plot "< awk '{if($3 == \"red\") print}' Data.csv" u 1:2 t "red" w p pt 2, \
     "< awk '{if($3 == \"green\") print}' Data.csv" u 1:2 t "green" w p pt 2, \
     "< awk '{if($3 == \"blue\") print}' Data.csv" u 1:2 t "blue" w p pt 2

you get this plot:

What awk does is simply check the third parameter of the data file and only print the line if it has some value: like red or blue.

You would also get rid of the palette with the gradient.

The script could be further improved by using gnuplot iterations.