3D graphs of probability density functions in MATLAB

Question

I want to create the 3D plot of the probability density functions of my variable. I have a matrix with dimensions 189x10000, where rows correspond to the time and columns are results of the simulation. Can somebody help me to create a density plot over time? I want my plot to look like this:

    A = [1:185]';  % substitute for date vector
    K = linspace( -20, 20, 100);
    f = zeros(185,100);
    xi = zeros(185,100);
    r = normrnd(0,1,[185,10000]);
     for i=1:185

        [f(i,:),xi(i,:)] = ksdensity(r(I,:));

     end
    a = figure;
    meshc(A, K', f')
    datetick('x', 'yyyy')
    view(85, 50)
    set(gca, 'YLim', [-15, 10])
    set(gca, 'XLim', [A(1), A(end)])
    xlabel('Time')

With this code I get this:

Racemic

each row of my Y_pred follows a different distribution for different time periods, and I want to show that evolution of density via 3D plot. The main issue is I don't get the density when I plot like this, but rather kind of uniformly scattered points in 3D form. – Racemic

Using random data you'd get random distributions of course. What is the problem in that plot? As mentioned, it looks like you actually want a waterfall() with a contour on the bottom, rather than a meshc which does both at once, but lashes the different distributions together along the y-axis. – Rustice

My bad, I was not plotting the density, just random numbers. I fixed it, but I still get this weird looking plot, its not as "refined " as the first one, I think the issue lies with the number of gridpoints I choose for ksdensity, Im not sure how to fix that. – Racemic

The plot actually looks fine. If you want a finer grid, i.e. more points, then use more points. Your actual data has 10-times as much, right? Otherwise this is as good as it gets; "improving" your plot to e.g. smooth over your data or something is more data-doctoring than science. – Rustice

Answer 1

1. Replace random numbers with density distribution.

2. If you want a finer grid, then use more points. Your actual data has 10-times as much, right? Otherwise this is as good as it gets; "improving" your plot, e.g. smooth over your data, is more data-doctoring than science.

Solution provided by Adriaan.

Mungo