3D scatter plot with 2D histogram at sides

Question

I would like to make a 3D scatterplot with with the 2D projections at each side. Something like the following:

The scatterplot has been created with:

fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
ax.scatter(x, y, z, c='r', marker='.')

And the projections with:

 h = np.np.histogram2d(x, y)
 plt.imshow(h, cmap='cubehelix_r', interpolation='none')

And they have been brought together with inkscape. How would I do this completely with matplotlib?

Answer 1

Use plot_surface():

ax.scatter(x,y,z, marker='.', c='r')

h, yedges, zedges = np.histogram2d(y, z, bins=50)
h = h.transpose()
normalized_map = plt.cm.Blues(h/h.max())

yy, zz = np.meshgrid(yedges, zedges)
xpos = min(x)-2 # Plane of histogram
xflat = np.full_like(yy, xpos) 

p = ax.plot_surface(xflat, yy, zz, facecolors=normalized_map, rstride=1, cstride=1, shade=False)

Repeat for the other 2 histograms.

A simpler method if you just want projections (but not histograms) is to add scatter plots with flattened data:

ax.scatter(x, y, z, c='r', marker='.')

xflat = np.full_like(x, min(ax.get_xlim()))
yflat = np.full_like(y, max(ax.get_ylim()))
zflat = np.full_like(z, min(ax.get_zlim()))

ax.scatter(xflat, y, z)
ax.scatter(x, yflat, z)
ax.scatter(x, y, zflat)

(Normally I'd just type x*0 + n to make a matching array filled with constant value, but np.full_like is more explicit)