I'm writing an interface for making 3D scatter plots in matplotlib, and I'd like to access the data from a python script. For a 2D scatter plot, I know the process would be:

import numpy as np
from matplotlib import pyplot as pltfig = plt.figure()
ax = fig.add_subplot(111)
h = ax.scatter(x,y,c=c,s=15,vmin=0,vmax=1,cmap='hot')
data = h.get_offsets()

With the above code, I know that data would be a (N,2) numpy array populated with my (x,y) data. When I try to perform the same operation for 3D data:

import numpy as np
from matplotlib import pyplot as plt
from mpl_toolkits.mplot3d import Axes3Dfig = plt.figure()
ax = Axes3D(fig)
h = ax.scatter(x,y,z,c=c,s=15,cmap='hot',vmin=0,vmax=1)
data = h.get_offsets()

The resulting data variable is still an (N,2) numpy array rather than a (N,3) numpy array. The contents of data no longer match any of my input data; I assume that data is populated with the 2D projections of my 3D data, but I would really like to access the 3D data used to generate the scatter plot. Is this possible?

Indeed, the coordinates obtained via get_offsets are the projected coordinates. The original coordinates are hidden inside the mpl_toolkits.mplot3d.art3d.Path3DCollection which is returned by the scatter in three dimensional axes. You would obtain the original coordinates from the ._offsets3d attribute. (This is a "private" attribute, but unfortunately the only way to retrieve this information.)

import numpy as np
from matplotlib import pyplot as plt
from mpl_toolkits.mplot3d import Axes3Dfig = plt.figure()
ax = Axes3D(fig)
x = [1,2,3,4]
y = [1,3,3,5]
z = [10,20,30,40]
c= [1,2,3,1]
scatter = ax.scatter(x,y,z,c=c,s=15,cmap='hot',vmin=0,vmax=1)
data = np.array(scatter._offsets3d).T
print(scatter)  # prints mpl_toolkits.mplot3d.art3d.Path3DCollection
print(data)# prints
# 
# [[  1.   1.  10.]
#  [  2.   3.  20.]
#  [  3.   3.  30.]
#  [  4.   5.  40.]]