8
$\begingroup$

I want to plot 6-dimensional data using parallel axes (https://en.wikipedia.org/wiki/Parallel_coordinates). The list looks like this:

    {{a1,b1,c1,d1,e1,f[a1,b1,c1,d1,e1]},{a2,b2,c2,d2,e2,f[a2,b2,c2,d2,e2]},{},...}

I need to draw six vertical axes with scales on them, connect the dots with lines, and color these lines according to the value of f[ai,bi,ci,di,ei], or the last coordinate.

A three dimensional example with two data would help me start, but I don't know how to do it.

Thank you.

$\endgroup$

2 Answers 2

11
$\begingroup$

Here's a start:

parallelPlot[dat_] := 
 ListPlot[dat, 
  Prolog -> 
   Table[Line[{{i, Min[dat]}, {i, Max[dat]}}], {i, 
     Length[dat[[1]]]}],
  PlotStyle -> 
   Table[Blend[{Red, Blue}, 
     Rescale[dat[[i, -1]], {Min[dat[[All, -1]]], Max[dat[[All, -1]]]}
      ]], {i, Length[dat]}],
  Joined -> True,
  Axes -> False
  ]

Test:

parallelPlot[RandomInteger[{-5, 5}, {10, 6}]]

enter image description here

Things that one can easily add but I don't have time now:

  1. Rescale each coordinate separately.
  2. Ticks, labels, etc.
  3. Colorbar

======= Better function, that also rescales ========

The Rescale function is what you should use for rescaling. Let's keep the minimum and maximum of each dimensions in the variables mins and maxs, and thus

{mins, maxs} = Transpose[{Min[#], Max[#]} & /@ Transpose[data]];
rescaledData = 
  Transpose@
   Table[Rescale[data[[All, i]], {mins[[i]], maxs[[i]]}], {i, Length@data[[1]]}];

We can then use mins and maxs to create tick labels at the bottom and top of each dimension. Here's the integrated code (I also added a Floor and Ceiling functions in the calculation of mins and maxs to make the bounds nicer):

parallelPlot[data_] := Block[{mins, maxs, rescaledData, n, ndim},
   n = Length[data];
   ndim = Length[data // First];
   {mins, maxs} = Transpose[{Floor@Min[#], Ceiling@Max[#]} & /@ Transpose[data]];
   rescaledData = 
    Transpose@
     Table[Rescale[data[[All, i]], {mins[[i]], maxs[[i]]}], {i, ndim}];
   ListPlot[rescaledData,
    Prolog -> Table[Line[{{i, 0}, {i, 1}}], {i, ndim}],
    PlotStyle -> Table[Blend[{Red, Blue}, rescaledData[[i, -1]]], {i, n}],
    Epilog -> {
     Table[Text[mins[[i]] // ToString, {i, 0}, {-1, 0}], {i, ndim}], 
     Table[Text[maxs[[i]] // ToString, {i, 1}, {-1, 0}], {i,ndim}]
    },
    Joined -> True, 
    Axes -> False];

To test, we generate 6-dimensional random data (here its 15 points), where each coordinate is randomly distributed between two random bounds.

d = Transpose@Table[
        RandomReal[{Min@#, Max@#} &@RandomInteger[{0, 10}, 2], 15],
    {6}]
parallelPlot@d

enter image description here

$\endgroup$
8
  • 2
    $\begingroup$ Like your avatar :) $\endgroup$
    – Silvia
    Aug 31, 2015 at 9:00
  • $\begingroup$ nice i.imgur.com/BFgoNIO.jpg $\endgroup$ Aug 31, 2015 at 9:53
  • 2
    $\begingroup$ @Silvia thx. Yours ain't bad neither :) $\endgroup$
    – yohbs
    Aug 31, 2015 at 9:55
  • 1
    $\begingroup$ Now that you've accepted my answer, I don't have any incentive to do so... Just kidding, I will add it later, maybe tonight. $\endgroup$
    – yohbs
    Aug 31, 2015 at 10:20
  • 1
    $\begingroup$ @user27670 I added the rescaling code. $\endgroup$
    – yohbs
    Aug 31, 2015 at 11:57
11
$\begingroup$

I took another stab at the problem to produce a self-contained function that generates axes as well as rescales the data. The function nplot is defined below. It takes a list of data as input

Clear[nplot]

nplot[data_?(ArrayQ[#, 2, NumericQ] &), ImageSize -> size_] := Module[
  {rescaled, plot, axes},
  rescaled = Transpose@(Rescale /@ Transpose@data);
  plot =
   ListLinePlot[
    rescaled, PlotRange -> {{1, Last@Dimensions@data}, Automatic},
    ImageSize -> size, PlotRangePadding -> None, Axes -> None
   ];
  axes =
   Map[
    ListPlot[
      data[[All, #]], PlotStyle -> None, 
      PlotRange -> {{1, Last@Dimensions@data}, Automatic},
      Axes -> {False, True}, PlotRangePadding -> None,
      ImageSize -> size, Background -> None, 
      AxesOrigin -> {#, Automatic}] &,
    Range[Last@Dimensions@data]
   ];
  Overlay[Flatten@{plot, axes}]
]

nplot[data_] := nplot[data, ImageSize -> Automatic]

Within nplot I generate a column-wise rescaled version of the data that is plotted using ListLinePlot; the axes of this plot are suppressed, because they will be replaced by more appropriate ones reflecting the non-rescaled value of the variables. I then use the existing ListPlot machinery to generate appropriately scaled and placed axes for the non-rescaled data, which are stored in axes within nplot. Finally, the rescaled plot and non-rescaled axes are overlaid and returned.

The function checks the nature of the input data (it must be at least a 2-dimensional array: a vector would result in a straight line across the plot and would be very uninteresting) and passes through values of the ImageSize option, if set by the user (otherwise a value of Automatic is used).


Here is what the function produces with some random input:

SeedRandom[123]
pts = RandomReal[{-10, 10}, {15, 6}];
nplot[pts, ImageSize -> Large]

n-dimensional plot output

$\endgroup$
3
  • $\begingroup$ Nice. I thought about using Overlay, but the problem is that this produces an object which is not a Graphics, and thus is not amenable to further manipulations. But if this is not an issue then that's probably a simpler approach. $\endgroup$
    – yohbs
    Aug 31, 2015 at 16:15
  • $\begingroup$ @yohbs That's very true. I have been trying to get around that limitation, but for now I'm still stuck with Overlay in order to harness a plotting function to auto-generate the axes. I have been looking around for pre-made functions that generate axes, or for access to the built-in ones, but no joy so far. $\endgroup$
    – MarcoB
    Aug 31, 2015 at 18:28
  • $\begingroup$ I opened a thread to see if someone solved it. You're welcome to join. $\endgroup$
    – yohbs
    Sep 15, 2015 at 20:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.