pyroaring 0.1.1

Fast and lightweight set for unsigned 32 bits integers.

This piece of code is a wrapper for the C library CRoaring. It provides a very efficient way to store and manipulate sets of (unsigned 32 bits) integers.

The wrapping used to be done with Ctypes. We recently switched to Cython for the following reasons:

• Much better performances for short function calls (e.g. addition/deletion of a single element).

• Easier installation, no need to install manually the C library, it is now distributed with PyRoaring.

• Extensibility, it will be easier to write efficient codes with Cython (e.g. some data structures based on roaring bitmaps).

If for some reason you wish to keep using the old version, based on Ctypes, use PyRoaring 0.0.7.

Requirements

• Environment like Linux and MacOS

• Python 2.7, or Python 3.3 or better

• A recent C compiler like GCC

• The Python package hypothesis (optional, for testing)

• The Python package Cython (optional, for compiling pyroaring from the sources)

Installation

To install pyroaring and the CRoaring library on your local account, use the following two lines:

pip install pyroaring --user # installs PyRoaringBitMap

To install them system-wide, use the following lines :

pip install pyroaring

Naturally, the latter may require superuser rights (consider prefixing the commands by sudo).

(If you want to use Python 3 and your system defaults on Python 2.7, you may need to adjust the above commands, e.g., replace pip by pip3 and python by python3.)

Manual compilation / installation

If you want to compile (and install) pyroaring by yourself, for instance to modify the Cython sources or because you do not have pip, follow these steps. Note that the Python package Cython is required.

Clone this repository.

git clone https://github.com/Ezibenroc/PyRoaringBitMap.git
cd PyRoaringBitMap

Build pyroaring locally, e.g. to test a new feature you made.

bash prepare_dist.sh
export PYROARING_CYTHON=1
python setup.py build_ext -i
python test.py # run the tests, optionnal but recommended

Install pyroaring (use this if you do not have pip).

python setup.py install # may require superuser rights, add option --user if you wish to install it on your local account

Package pyroaring.

python setup.py sdist
pip install dist/pyroaring-0.0.7.tar.gz # optionnal, to install the package

Utilization

First, you can run the tests to make sure everything is ok:

pip install hypothesis --user
python test.py

You can use a bitmap nearly as the classical Python set in your code:

from pyroaring import BitMap
bm1 = BitMap()
bm1.add(3)
bm1.add(18)
bm2 = BitMap([3, 27, 42])
print("bm1       = %s" % bm1)
print("bm2       = %s" % bm2)
print("bm1 & bm2 = %s" % (bm1&bm2))
print("bm1 | bm2 = %s" % (bm1|bm2))

Output:

bm1       = BitMap([3, 18])
bm2       = BitMap([3, 27, 42])
bm1 & bm2 = BitMap([3])
bm1 | bm2 = BitMap([3, 18, 27, 42])

Benchmark

The built-in set is compared with this Python wrapper of CRoaring (designated as pyroaring in the following) and a Cython implementation of Roaring bitmaps (designated as cyroaring in the following).

Quick benchmarks for common operations

The script quick_bench.sh measures the time of different set operations. It uses sets initialized to range(b, 100000000, 8) with b equal to 0 or 1. It is far from being exhaustive, but rather a quick overview of how the three classes compare to each other.

Operation	Pyroaring	Cyroaring	set
Empty constructor	0.000134	0.000154	7.75e-05
Range constructor	4.51	4.15	756
List constructor	153	124	584
Element addition	7.09e-05	7.06e-05	6.5e-05
Test for membership	3.03e-05	3.28e-05	2.6e-05
Conversion to list	513	486	146
Equality test	1.47	1.35	289
Subset test	1.54	1.46	283
Union	3.18	3.4	811
Intersection	2.58	2.52	132
Symetric difference	3.13	3.19	927
Pickle dump & load	17.4	17.3	1.29e+03
Selection	0.00754	0.00135	NA
Slice	614	2.88e+03	NA

Complete benchmark for the union

The performances of the union operation have been measured more carefully. Full results can be found here.

To sum up, both Roaring bitmap implementations are several orders of magnitude faster than the built-in set, regardless of the density of the data.

For sparse data, pyroaring is faster than cyroaring, for very dense data cyroaring is faster. Otherwise, they are similar.