Matplotlib: "Unknown projection '3d'" error

I encounter the same problem, and @Joe Kington and @bvanlew's answer solve my problem.

but I should add more infomation when you use pycharm and enable auto import.

when you format the code, the code from mpl_toolkits.mplot3d import Axes3D will auto remove by pycharm.

so, my solution is

from mpl_toolkits.mplot3d import Axes3D
Axes3D = Axes3D  # pycharm auto import
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')

and it works well!

Just to add to Joe Kington's answer (not enough reputation for a comment) there is a good example of mixing 2d and 3d plots in the documentation at http://matplotlib.org/examples/mplot3d/mixed_subplots_demo.html which shows projection='3d' working in combination with the Axes3D import.

from mpl_toolkits.mplot3d import Axes3D
...
ax = fig.add_subplot(2, 1, 1)
...
ax = fig.add_subplot(2, 1, 2, projection='3d')

In fact as long as the Axes3D import is present the line

from mpl_toolkits.mplot3d import Axes3D
...
ax = fig.gca(projection='3d')

as used by the OP also works. (checked with matplotlib version 1.3.1)

Import mplot3d whole to use "projection = '3d'".

Insert the command below in top of your script. It should run fine.

from mpl_toolkits import mplot3d

First off, I think mplot3D worked a bit differently in matplotlib version 0.99 than it does in the current version of matplotlib.

Which version are you using? (Try running: python -c 'import matplotlib; print matplotlib."__version__")

I'm guessing you're running version 0.99, in which case you'll need to either use a slightly different syntax or update to a more recent version of matplotlib.

If you're running version 0.99, try doing this instead of using using the projection keyword argument:

import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import axes3d, Axes3D #<-- Note the capitalization! 
fig = plt.figure()

ax = Axes3D(fig) #<-- Note the difference from your original code...

X, Y, Z = axes3d.get_test_data(0.05)
cset = ax.contour(X, Y, Z, 16, extend3d=True)
ax.clabel(cset, fontsize=9, inline=1)
plt.show()

This should work in matplotlib 1.0.x, as well, not just 0.99.