Retrieving the numer of bytes consumed by a SciPy sparse matrix

Question

Let's say I want to monitor the memory occupied by my SciPy sparse matrix mat. In NumPy I would have exploited the nbytes attribute, but in SciPy it seems there is nothing like that. How can I retrieve this information?

Answer 1

I have sparse matrix X

In [605]: X
Out[605]: 
<100x100 sparse matrix of type '<class 'numpy.float64'>'
    with 1000 stored elements in Compressed Sparse Row format>

getsizeof doesn't tell me anything useful

In [606]: import sys
In [607]: sys.getsizeof(X)
Out[607]: 28

The sparse data and indices are, for a csr matrix stored in 3 arrays:

In [612]: X.data.nbytes
Out[612]: 8000
In [613]: X.indices.nbytes
Out[613]: 4000
In [614]: X.indptr.nbytes
Out[614]: 404

So roughly the total space is the sum of those values.

For coo format

In [615]: Xc=X.tocoo()
In [616]: Xc.data.nbytes
Out[616]: 8000
In [617]: Xc.row.nbytes
Out[617]: 4000
In [618]: Xc.col.nbytes
Out[618]: 4000

We could calculate those values from shape, dtype and nnz; e.g. 8 bytes * 1000, 4bytes * 1000, 4bytes * X.shape[0], etc.

Other formats require knowledge of their data storage methods (e.g. lil, dok, etc).