Remove all zero rows and columns from symmetric numpy matrix

Question

I have this symmetric matrix:

>>> X = np.array([[2,0,1,0],[0,0,0,0],[1,0,1,0], [0,0,0,0]])

array([[2, 0, 1, 0],
       [0, 0, 0, 0],
       [1, 0, 1, 0],
       [0, 0, 0, 0]])

What's the most concise way to remove rows and columns that are allzero? Such that I get:

array([[2, 1],
       [1, 1]])

Is there a term for this operation, or a one-liner to do it?

I tried the following, but it did not work:

mask = np.all(X == 0, axis=1)
X[not(mask),not(mask)]
>>> Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

Answer 1

rows = np.argwhere(np.sum(X, axis=0) == 0).flatten()
cols =  np.argwhere(np.sum(X, axis=1) == 0).flatten()
np.delete(np.delete(X, rows, axis=0), cols, axis=1)

Answer 2

Have you had a look what np.all returns? It's an array, so you can't simply 'not' it. Rather than that, you could try np.any ('not all = any' - good old de Morgan):

mask = np.any(X, axis=1)

Then you can get your 'filtered' X like this:

X[mask][:, mask]

Note that once you do that, you may potentially end up with new zeroes-filled rows and columns, so you might want to repeat this operation several times.

Answer 3

This appears to work but seems dumb:

>>> mask = np.all(X == 0, axis=1)
>>> X = X[~mask, :]
>>> X = X[:, ~mask]
>>> X
array([[2, 1],
       [1, 1]])