ruby gnuplot heatmap for Self Organizing Map outputs

Question

I'm doing some experiments on SOM in ruby. After getting best match unit, my output is looking like that : x coordinate, y coordinate and distance from the best match unit

I have 200 lines output.

I want to make an heatmap for this output but getting in troubles with gnuplot.

the code I actually use :

Gnuplot.open do |gp|
  Gnuplot::Plot.new( gp ) do |plot|
    plot.title  "heatmap"
    plot.xlabel "x"
    plot.ylabel "y"
    plot.xrange "[0:50]"        
    plot.yrange "[0:50]"        
    plot.terminal "png"
    plot.output "myheatmap.png"
    plot.set "pm3d map"
    plot.set "palette rgbformula -7,2,-7" # for green heatmap
    plot.set "autoscale fix"       
    plot.cbrange "[0.0:0.036]" # range of distance
    plot.cblabel "Score"     
    plot.unset "cbtics"
    plot.data << Gnuplot::DataSet.new( [x,y,dist] ) do |ds|
      ds.with = "image"
      ds.notitle
    end
  end
end

x and y array contains integer (for coordinates) and dist array contains float values.

I've already search here and on gnuplot site. Green color used is inspired from here : Gnuplot: how to write the z values in a heatmap plot

I success in plotting distance with points, but I need more visual plot with heatmap.

Help will be appreciate thanks

edit : Hi Christoph, my data looks like the few lines below :

24,15,1.1214532012163798e-13,
41,41,2.088613932696844e-13,
4,10,1.485599551044706e-13,
0,20,7.981851602311569e-14,
6,46,1.1231898879790176e-13,
8,24,1.2844471889344152e-13,
11,24,2.3794505905958835e-13,
3,16,0.015633285670666745,
3,46,1.238425800407315e-13,
4,20,1.2695729760609708e-13,

I map each column into the appropriate type (int for the first two and float for the last). x and y represent the node coordinates and the last column represent distance between input and the best match unit. I want to plot an heatmap (50x50) and colorize each node depending of the distance (white for short distance, green darker for the biggest distance).

The main problem is that I didn't success to plot what I want with the code pasted below. I guess I can't pasted screen-shot since I've not enough post..

edit 2 : I've also try to round distance column, no changes

edit 3 : I finally transform my output (3 vectors : x, y, distance) into a regular grid.

I've created a two dimensional array filled with 1.0 :

ntab = Array.new(50) { |i| Array.new(50) { |i| 1.0 }}

then I look for corresponding index in x and y vectors, that indicates the node is present in my vectors, and fill the ntab node with corresponding distance value :

(1..50).each do |xv|
  (1..50).each do |yv|
    resx = x.each_with_index.select { |x, idx| x == xv}
    resy = y.each_with_index.select { |y, idy| y == yv}
    resx.each do |row|
      next if row.empty?
      tmp = [yv,row[1]]
      if resy.include?(tmp)
      then 
        ntab[xv-1][yv-1] = dist[row[1]] 
      end
    end
  end
end

I've now a perfect grid (50x50) from which I can plot distance value. I will edit my question with final working code if I find it :)

Thanks Christoph for advices about gnuplot

So whats your actual Problem? *What* doesn't work? How dies your data file look like? Probably you are just missing an empty line between blocks of equal x- values. — Christoph, Feb 21 '15 at 16:29
OK. The main point is, that you need a regular grid to plot a heat map. There is `dgrid3d` to create a grid for you, but that's most probably not what you want because it interpolates your data to the point of a regular grid. If your nodes don't really make a grid, or shouldn't be interpolated to create a continuous surface, you might want to use a different plotting style, like circles — Christoph, Feb 21 '15 at 23:01
If you use `plot ... with image`, then you get the colors you actually want. Using `splot ... with pm3d` determines the color of a single quadrangle by interpolating the values of its four corners (see pm3d's option `corners2color` about how this interpolation can be chosen). — Christoph, Feb 23 '15 at 08:18

score 0 · Accepted Answer · edited Nov 22 '15 at 20:49

the result I finally have isn't really what I want but it do the trick to analyze SOM clusters

fill node array with max value of distance and then plot with the code below (I didn't put link because I need also 10 reputations points to post more than 2 links :/)

Gnuplot.open do |gp|
  Gnuplot::SPlot.new( gp ) do |plot|
    plot.pm3d
    plot.hidden3d
    plot.palette 'defined (   0 "black", 51 "blue", 102 "green", 153 "yellow", 204 "red", 255 "white" )'
    plot.data << Gnuplot::DataSet.new( ntab ) do |ds|
      ds.with = 'lines'
      ds.matrix = true
    end
  end
end

Result

ruby gnuplot heatmap for Self Organizing Map outputs

1 Answers1