FileIO Usage

Example source code can be found in ./examples/io.py.

File I/O is an important way to keep your data. PyVSparse supports reading/writing files in our own format.

csc_mat = sp.sparse.random(3, 3, format='csc', dtype=np.float32, density=1)

# Write the matrix to a file
vcsc = PyVSparse.VCSC(csc_mat)
vcsc.write('vcsc_mat') # This will automatically append .vcsc to the end, if not already there

# Read the matrix from the file
vcsc2 = PyVSparse.VCSC('vcsc_mat.vcsc')

Importantly, the written file will maintain the information of the original matrix. So if the written matrix uses 2 byte index types, the read matrix will also use 2 byte index types. No customization or extra construction paramters are needed.

vcsc2 = PyVSparse.VCSC(csc_mat, indexType=np.uint16)
vcsc2.write('vcsc_mat2')

vcsc3 = PyVSparse.VCSC('vcsc_mat2.vcsc')
print(vcsc3.indexType == np.uint16) # True

The same can be done with IVCSC matrices, however, IVCSC matrices do not have index type parameters.

ivcsc = PyVSparse.IVCSC(csc_mat)
ivcsc.write('ivcsc_mat')

ivcsc2 = PyVSparse.IVCSC('ivcsc_mat.ivcsc')
print(ivcsc2 == ivcsc2) # True

The files written are compatible in Python and C++.

npz File I/O

For convienece, we also support reading and writing npz files. Writing these files are done through the SciPy library. This is essentially a wrapper around SciPy’s load_npz function, as it still constructs the CSC matrix.

import os
try:
    sp.sparse.save_npz("test.npz", csc_mat)
    newVCSC = PyVSparse.VCSC("test.npz")
finally:
    os.remove("test.npz")
print(newVCSC.tocsc().toarray() == csc_mat.toarray()) # True

try:
    sp.sparse.save_npz("test.npz", csc_mat)
    newIVCSC = PyVSparse.IVCSC("test.npz")
finally:
    os.remove("test.npz")
print(newIVCSC.tocsc().toarray() == csc_mat.toarray()) # true

If you have a .npz file, but the format is BSR or some other incompatible format, then you may use the secondary parameter to specify what format to convert the written matrix to. By default, this is set to CSC.

import os
try:
    bsr_mat = csc_mat.tobsr()
    sp.sparse.save_npz("bsr.npz", csc_mat)
    newVCSC = PyVSparse.VCSC("bsr.npz", secondary="csr")
finally:
    os.remove("bsr.npz")
print(newVCSC.tocsc().toarray() == csc_mat.toarray()) # True

try:
    sp.sparse.save_npz("bsr.npz", csc_mat)
    newIVCSC = PyVSparse.IVCSC("bsr.npz", secondary="coo")
finally:
    os.remove("bsr.npz")
print(newIVCSC.tocsc().toarray() == csc_mat.toarray()) # true