CODEWITHSUNDEEP
pythonnumpymatrixscipysparse-matrix
dshepherd
dshepherd

Reputation: 5407

Is it possible to create a dummy sparse matrix with no rows or columns in NumPy?

I'm working with arrays of sparse matrices using SciPy, and sometimes some of the matrices have no rows or columns (that is, shape = (0,0)). I would like my various other functions to just ignore these matrices, or do some other smart default action.

It seems I can't use any of the sparse matrix formats, because they complain about "invalid shape". For some reason matrix([]) and matrix([[]]) have shape = (1,0), so these are not much good (actually, is that a bug? It makes little sense...).

I know that I could use None, [], array([]) or something similar. However standard matrix manipulation functions will fail when handed these, meaning I would need to scatter

if block is not None:

or similar throughout my code.

Is there a matrix class that I've missed which will allow me to handle these cases gracefully?

For anyone curious: this comes out of chopping up a finite element Jacobian matrix into blocks by the type of variable. Some of these variables may be pinned (fixed to a constant value), and so have no derivatives. Hence no entries in the Jacobian.

Upvotes: 2

Views: 1353

Answers (4)

shyamrag cp
shyamrag cp

Reputation: 102

You can try below codes and it is easy.

import numpy as np np.zeros((0,0))

Upvotes: 0

Warren Weckesser
Warren Weckesser

Reputation: 114946

There are many ways to create a 0x0 sparse matrix. For example,

In [16]: from scipy.sparse import csr_matrix, coo_matrix, dok_matrix, eye

In [17]: csr_matrix(([], [[],[]]), shape=(0,0))
Out[17]: 
<0x0 sparse matrix of type '<type 'numpy.float64'>'
with 0 stored elements in Compressed Sparse Row format>

In [18]: coo_matrix(([],[[],[]]), shape=(0,0))
Out[18]: 
<0x0 sparse matrix of type '<type 'numpy.float64'>'
with 0 stored elements in COOrdinate format>

In [19]: dok_matrix((0,0))
Out[19]: 
<0x0 sparse matrix of type '<type 'numpy.float64'>'
with 0 stored elements in Dictionary Of Keys format>

In [20]: eye(0)
Out[20]: 
<0x0 sparse matrix of type '<type 'numpy.float64'>'
with 0 stored elements (1 diagonals) in DIAgonal format>

In [21]: csr_matrix(np.zeros((0, 0)))
Out[21]: 
<0x0 sparse matrix of type '<type 'numpy.float64'>'
with 0 stored elements in Compressed Sparse Row format>

You said you were chopping up a Jacobian matrix. If you index a sparse matrix with slices that have zero length, you'll get a 0x0 result:

In [22]: a = np.random.randint(0, 2, size=(10,10))

In [23]: m = csr_matrix(a)

In [24]: m[3:3, 3:3]
Out[24]: 
<0x0 sparse matrix of type '<type 'numpy.int64'>'
with 0 stored elements in Compressed Sparse Row format>

(I'm using scipy 0.13.2.)

It is possible that some sparse operations will not handle a 0x0 matrix properly, and that might be a bug. If you have some examples, could you add them to your question?

Upvotes: 1

Eelco Hoogendoorn
Eelco Hoogendoorn

Reputation: 10769

matrix([[]]) -> shape = (1,0) makes perfect sense; you gave it one row, with zero columns; how is numpy supposed to guess what you really wanted was zero rows instead?

All array creation functions such as np.zeros((0,0)) work for your purposes.

Upvotes: 1

YXD
YXD

Reputation: 32521

Yes. You can do

from scipy import sparse
result = sparse.eye(0)
# result.shape = (0, 0)

Upvotes: 3

Related Questions