PyRSB: a Cython-based Python interface to librsb
Project description
PyRSB
librsb is a high performance sparse matrix library implementing the Recursive Sparse Blocks format, which is especially well suited for multiplications in iterative methods on huge symmetric sparse matrices.
PyRSB is a Cython-based Python interface to librsb.
On multicore machines, PyRSB can be several times faster than e.g. scipy.sparse.csr_matrix()
.
For an example how to invoke it with no overhead, see the advanced example.
So far, PyRSB is a prototype tested on Linux only. The librsb library offers much more, and can make PyRSB much more powerful. Prospective users and collaborators feedback are sought; please contact me to feedback and help.
Features
The following functionality is implemented:
- Initialization with
rsb.rsb_matrix()
styled asscipy.sparse.csr_matrix()
. - Conversion from
scipy.sparse.csr_matrix()
. - Multiplication by vector/multivector.
- Rows/columns through
nr=a.shape()[0]
/nr=a.shape()[1]
, ornr()
/nc()
. find()
,find_block()
,tril()
,triu()
,shape()
,nnz
.print
'able.- PyRSB-Specific:
autotune()
,do_print()
. - load from a Matrix Market file, e.g.
rsb.rsb_file_mtx_load(bytes(filename,encoding='utf-8'))
Build and Use
- If you have librsb installed:
make
shall build and test. - Make sure you have
cython
,scipy
,numpy
. installed. - If you want to install librsb on Ubuntu or Debian:
sudo apt-get install librsb-dev
shall suffice. Other operating systems have librsb, too. Please check yours. Or check librsb's web site. - If you want the
Makefile
to build librsb (in this directory):make all-local
will attempt downloading librsb-1.2.0.9 from the web and building it here before building pyrsb. If the file is in place, it won't download it a second time. After that,make local-librsb-pyrsb
(ormake lp
) will build pyrsb using local librsb, then run it. make test
will test benchmark code usingtest.py
(to compare speed to SciPy)make b
will also produce graphs (requiresgnuplot
)
Example Usage
# Example: demo1.py
"""
pyrsb demo
"""
import numpy
import scipy
from scipy.sparse import csr_matrix
from rsb import rsb_matrix, rsb_dtype
V = [11.0, 12.0, 22.0]
I = [0, 0, 1]
J = [0, 1, 1]
c = csr_matrix((V, (I, J)))
print(c)
# several constructor forms, as with csr_matrix:
a = rsb_matrix((V, (I, J)))
a = rsb_matrix((V, (I, J)), [3, 3])
a = rsb_matrix((V, (I, J)), sym="S") # symmetric example
print(a)
a = rsb_matrix((4, 4))
a = rsb_matrix(c,dtype=rsb_dtype)
nrhs = 1 # set to nrhs>1 to multiply by multiple vectors at once
nr = a.shape[0]
nc = a.shape[1]
order = "F"
x = numpy.empty([nc, nrhs], dtype=rsb_dtype, order=order)
y = numpy.empty([nr, nrhs], dtype=rsb_dtype, order=order)
x[:, :] = 1.0
y[:, :] = 0.0
print(a)
print(x)
print(y)
# import rsb # import operators
# a.autotune() # makes only sense for large matrices
y = y + a * x
# equivalent to y=y+c*x
print(y)
del a
Example Advanced Usage
# Example: demo2.py
"""
pyrsb demo
"""
import numpy
import scipy
from scipy.sparse import csr_matrix
from rsb import rsb_matrix, rsb_dtype
V = [11.0, 12.0, 22.0]
I = [0, 0, 1]
J = [0, 1, 1]
a = rsb_matrix((V, (I, J)))
nrhs = 4 # set to nrhs>1 to multiply by multiple vectors at once
nr = a.shape[0]
nc = a.shape[1]
# Choose Fortran or "by columns" order here.
order = "F"
x = numpy.empty([nc, nrhs], dtype=rsb_dtype, order=order)
y = numpy.empty([nr, nrhs], dtype=rsb_dtype, order=order)
x[:, :] = 1.0
y[:, :] = 0.0
print(a)
print(x)
print(y)
# Autotuning example: use it if you need many multiplication iterations on huge matrices (>>1e6 nonzeroes).
# Here general (nrhs=1) case:
a.autotune()
# Here with all the autotuning parameters specified:
a.autotune(1.0,0,1,2.0,ord('N'),1.0,nrhs,ord('F'),1.0,False)
# Inefficient: reallocate y
y = y + a * x
# Inefficient: reallocate y
y += a * x
# Equivalent but more efficient: don't reallocate y
a._spmm(x,y)
print(y)
del a
License
GPLv3+
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distributions
Hashes for pyrsb-0.2.20210226-cp39-cp39-manylinux2010_x86_64.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 9d114b1df3f5424455aeec17149dbb81d5c78f8c7ce63963ef32445a8d2fb34a |
|
MD5 | 8047195c539f3a8b421b2ad0ac6cfa44 |
|
BLAKE2b-256 | cc4eec38979cebc8f94dd8db3133113ddcdc9e345414e389339347719fc72b8e |
Hashes for pyrsb-0.2.20210226-cp38-cp38-manylinux2010_x86_64.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | bd46b9506ab0ffa87852696b1711d28e7264bf77a0fdd22fdbaf4601d5c1c5b3 |
|
MD5 | b8be2a979fe4a1ff0888eb0c5c289187 |
|
BLAKE2b-256 | 99562abad30385fa6e2ddcaf963d6e22e54d864d64e6ef19d36049ff1be0b4a0 |
Hashes for pyrsb-0.2.20210226-cp37-cp37m-manylinux2010_x86_64.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | f54464a4df4d56f4fbbd6a5fd9e9a28bc6ed016de66875c4b1586307da47832b |
|
MD5 | 057eb8ba120dcf64bd150b417a00f5bb |
|
BLAKE2b-256 | 20f3455164ce3923b678824056a7b0a6f4cecacf5da4d3ce76a083b0245b0eb8 |
Hashes for pyrsb-0.2.20210226-cp36-cp36m-manylinux2010_x86_64.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 08c69724883d631e87fad4bfa1fc19d9c1348f9a822b4700b574f7d64da1552b |
|
MD5 | 84651990e9b5a592b1e255705415974e |
|
BLAKE2b-256 | d5f736de04174ab300e736c89fc6ef8b3f75eca8b2461ff6941e839d1e7a2151 |