Wrong axis value in correlation plot

Hi, I used correlation plot(corrplot) and getting the following graph. And it is obvious the var2 is bounded by -1 from the two scatter plot and as well as from the data. But the second histogram starts from a starting value much smaller value. I'm not sure is it a bug or my setting problem. Actually I did not do any settings, this is the default output from corrplot.

[the cyclist]

Can you upload your data and code? This would make your problem easier to diagnose.

[JT Yang]

Hi, it is simple two columns of data as in the attachment.

command used:

load('matlab.mat')

corrplot(MCMC)

MATLAB R2019A linux version

Answer 1

[the cyclist]

I don't have the Econometrics Toolbox, so I can't confirm myself. But if that is really the only thing you did, then I think this is either a bug or a very unconventional choice on the part of Mathworks. I did this same style plot in R (using the psych library), and the output as you expect it -- with labeling of the figures aligned for both the histogram and the scatterplot. I can't think of a reason to not do it that way.

So, if you are sure that you have not adjusted the scatterplot axes after running the above commands, then I would submit a bug report, including your data and code.

[JT Yang]

Hi, thanks cyclist, I am also testing with Python matplotlib, it seems fine to plot the same type plot.

I'm not sure whether it is a bug or design preference. But I think it is good to share.

Answer 2

[dpb]

It's a confusing labeling problem -- the ticks aren't actually the variables on the plot for the histogram but for the scatter plots -- the axes values are for the scatter plot in the upper right quadrant which is scatter(x(:,2),x(:,1)) while the lower left is scatter(x(:,1),x(:,2)).

So, the x-axes limits are for var2 which are bounded at -1; the labels have nothing to do with the histograms; only the scatter plots. TMW just placed the x-axis labels for all the suplots at the bottom instead of keeping them with their respective suplot locations.

If create a poor-man's copy...

subplot(2,2,1)
hist(MCMC(:,1),100)
subplot(2,2,2)
scatter(MCMC(:,2),MCMC(:,1))
ylim([0 0.025])
xlim([-1.105 -0.7])
subplot(2,2,3)
scatter(MCMC(:,1),MCMC(:,2))
xlim([0 0.025])
ylim([-1.105 -0.7])
subplot(2,2,4)
hist(MCMC(:,2),100)
xlim([-1.05 -0.75])

one gets

so, while perhaps confusing, looks like the data are in fact correct...I don't have the Econometrics TB so can't go in and query what the actual limits are on each subplot() but doing so will make things clear, I expect. I fiddled around to roughly approximate the appearance on the LRH one...that took some doing...what the logic is that created the actual limits would be interesting to go look at, perhaps.

[JT Yang]

Hi, dpb, thanks a lot for your anwser. Yes, I think the histogram is correct. It is the axis labeling only for the scatter plot not for the histogram.

Personally I feel without aligning the histogram with scatter plot will confuse audiance if such graph is to be used for presentation or report. Or at least I feel it is more reasonable to have them aligned.

I tried to find any options to align the axis, but no luck for now.

[dpb]

As Cyclist says, this is a quality of implementation problem if not an outright bug so is worthy of submitting a service request on.

I presume they didn't do anything more exotic that create a bunch of specialized subplot() with position adjusted and axes hidden...as suggested earlier, if you'll get the axes handles from the figure created, you can directly query the xlim property for each subplot() and determine precisely what those values are.

Or, if it is an m-file, you could go read the source and try to decipher how they did set the x-limit property for each.

That would make it absolutely clear whether my supposition is, in fact, correct (altho I'm quite confident it is, there's always a chance there's a more significant bug)

[JT Yang]

Hi, dpb,

Thank you for your input. I tried out to edit the axis limit. I realize I can edit the xlim for both histogram. But axis limit for scatter plot is fixed as default and not shown. I can shift the xlim of histogram visually to match the scatter graph. It looks a lot better but not exact.

[dpb]

Save the .fig file and attach it. Even w/o the TB, should be able to open the .fig file and poke around inside...

[JT Yang]

Hi, dpb, many thanks for your help. I've attached the fig file for your reference.

[dpb]

Wowsers!!! They went all out...

>> openfig('corrplot.fig')
Warning: Unable to load instances of class matlab.ui.controls.AxesToolbar into a heterogeneous array.  The definition of
...
ans = 
  Figure (2) with properties:

      Number: 2
        Name: ''
       Color: [0.9400 0.9400 0.9400]
    Position: [675 549 570 420]
       Units: 'pixels'

  Show all properties
>> hF=gcf;
>> hA=hF.Children
hA = 
  7×1 Axes array:

  Axes    (PlotMatrixHistAx)
  Axes    (PlotMatrixHistAx)
  Axes    (PlotMatrixScatterAx)
  Axes    (PlotMatrixScatterAx)
  Axes    (PlotMatrixScatterAx)
  Axes    (PlotMatrixScatterAx)
  Axes    (CorrPlot)
>> xl=get(hA,'XLim');
>> xl{:}
ans =
    0.0032    0.0280
ans =
   -1.0300   -0.7700
ans =
    0.0051    0.0261
ans =
   -1.1193   -0.7334
ans =
    0.0051    0.0261
ans =
   -1.1193   -0.7334
ans =
     0     1
>> 

Looks like they created an axes for each quadrant for starters and plotted scatter and then overlaid two additional axes for the two histograms on top...instead of just a simple four-axes subplot() thing.

But, you can see where they left the ranges after done by default...

I confirmed that axes 2 and 6 are the LRH corner by

text(hA(6),-1,-0.8,'SIX')
text(hA(2),-1,7000,'TWO')
hA(6).Color='none';
hA(2).Color='none';

which also reveals they drew a line on the scatter axes that seems to traverse the range of the scatter plot data--I don't understand the purpose for it to then hide it...

Curiouser and curiouser... :)

ADDENDUM:

It appears the line is plot() of the scatter data sorted by increasing x...for what purpose still escapes me--they could have easily found the ranges by simply using min()/max() unless were too lazy and letting the autoranger default the axis limits--but even that could be just the two extrema; why duplicate the whole data set yet again??? But no, that isn't it because the limits aren't even as would otherwise be and

>> hA(6).XLimMode
ans =
    'manual'
>> 

confirms that.

Bizarre.

[JT Yang]

Hi, dpb. You are superb. I've never dive down so deep into plotting. Now I know how to get the xlim for all graphs. That will be helpful for get around.

[dpb]

One additional note: The line for the diagonal axes that isn't shown is just the variable plotted against itself--the 'ii' case so it is simply a 45-degree line of the range of the variable. The just didn't bother to fix up the logic to not plot the i==j case but did all and then overwrote the diagonal with the histograms not bothering to delete() the extra axes/line.

Answer 3

[Amin Salehi]

Yup, I see the same bug (different dataset) the histogram x-axis are wrong.