I have a sparse matrix B, I want to get the sparse matrix A from B by summation of all rows in the first column, then dividing the first column by '2', and making the other columns zero.
from numpy import array
from scipy import csr_matrix
row = array([0,0,1,2,2,2])
col = array([0,2,2,0,1,2])
data = array([1,2,3,4,5,6])
B = csr_matrix( (data,(row,col)), shape=(3,3) )
A = B.copy()
A = A.sum(axis=1)/2
# A shape becomes 1 x 3 instead of 3 x 3 here!
I think this could be approached in several ways. Your's is fine.
In [275]: from scipy.sparse import csr_matrix
...:
...: row = np.array([0,0,1,2,2,2])
...: col = np.array([0,2,2,0,1,2])
...: data = np.array([1,2,3,4,5,6.]) # make float
...:
...: B = csr_matrix( (data,(row,col)), shape=(3,3) )
In [276]: A = B.copy()
In [277]: A
Out[277]:
<3x3 sparse matrix of type '<class 'numpy.float64'>'
with 6 stored elements in Compressed Sparse Row format>
The assignment works:
In [278]: A[:,0] = A.sum(axis=1)/2
/usr/local/lib/python3.6/dist-packages/scipy/sparse/_index.py:126: SparseEfficiencyWarning: Changing the sparsity structure of a csr_matrix is expensive. lil_matrix is more efficient.
self._set_arrayXarray(i, j, x)
In [279]: A[:,1:] = 0
/usr/local/lib/python3.6/dist-packages/scipy/sparse/_index.py:126: SparseEfficiencyWarning: Changing the sparsity structure of a csr_matrix is expensive. lil_matrix is more efficient.
self._set_arrayXarray(i, j, x)
In [280]: A
Out[280]:
<3x3 sparse matrix of type '<class 'numpy.float64'>'
with 9 stored elements in Compressed Sparse Row format>
In [283]: A.eliminate_zeros()
In [284]: A
Out[284]:
<3x3 sparse matrix of type '<class 'numpy.float64'>'
with 3 stored elements in Compressed Sparse Row format>
In [285]: A.A
Out[285]:
array([[1.5, 0. , 0. ],
[1.5, 0. , 0. ],
[7.5, 0. , 0. ]])
The efficiency warning is mainly intended to discourage iterative or repeated assignments. I think that for one-time actions like this they can be ignored.
Or if we start with an all-zero A:
In [286]: A = csr_matrix(np.zeros(B.shape)) # may be better method
In [287]: A[:,0] = B.sum(axis=1)/2
/usr/local/lib/python3.6/dist-packages/scipy/sparse/_index.py:126: SparseEfficiencyWarning: Changing the sparsity structure of a csr_matrix is expensive. lil_matrix is more efficient.
self._set_arrayXarray(i, j, x)
In [288]: A
Out[288]:
<3x3 sparse matrix of type '<class 'numpy.float64'>'
with 3 stored elements in Compressed Sparse Row format>
Alternatively that column sum matrix could be used to construct A directly, using the same style of inputs as used to define B:
In [289]: A1 = B.sum(axis=1)/2
In [290]: A1
Out[290]:
matrix([[1.5],
[1.5],
[7.5]])
In [296]: row = np.arange(3)
In [297]: col = np.zeros(3,int)
In [298]: data = A1.A1
In [299]: A = csr_matrix((data, (row, col)), shape=(3,3))
In [301]: A
Out[301]:
<3x3 sparse matrix of type '<class 'numpy.float64'>'
with 3 stored elements in Compressed Sparse Row format>
In [302]: A.A
Out[302]:
array([[1.5, 0. , 0. ],
[1.5, 0. , 0. ],
[7.5, 0. , 0. ]])
I don't know which approach is fastest. Your sparse.hstack looks nice, though under the covers, hstack is building the row,col,data arrays from the coo formats, and making a new coo_matrix. While it is reliable, it's not particularly streamlined.