README.rst

PyProbables

pyprobables is a pure-python library for probabilistic data structures. The goal is to provide the developer with a pure-python implementation of common probabilistic data-structures to use in their work.

To achieve better raw performance, it is recommended supplying an alternative hashing algorithm that has been compiled in C. This could include using the md5 and sha512 algorithms provided or installing a third party package and writing your own hashing strategy. Some options include the murmur hash mmh3 or those from the pyhash library. Each data object in pyprobables makes it easy to pass in a custom hashing function.

Read more about how to use Supplying a pre-defined, alternative hashing strategies or Defining hashing function using the provided decorators.

Installation

Pip Installation:

$ pip install pyprobables

To install from source:

To install pyprobables, simply clone the repository on GitHub, then run from the folder:

$ python setup.py install

pyprobables supports python 3.6 - 3.11+

For python 2.7 support, install release 0.3.2

$ pip install pyprobables==0.3.2

API Documentation

The documentation of is hosted on readthedocs.io

You can build the documentation locally by running:

$ pip install sphinx
$ cd docs/
$ make html

Automated Tests

To run automated tests, one must simply run the following command from the downloaded folder:

$ python setup.py test

Quickstart

Import pyprobables and setup a Bloom Filter

from probables import BloomFilter
blm = BloomFilter(est_elements=1000, false_positive_rate=0.05)
blm.add('google.com')
blm.check('facebook.com')  # should return False
blm.check('google.com')  # should return True

Import pyprobables and setup a Count-Min Sketch

from probables import CountMinSketch
cms = CountMinSketch(width=1000, depth=5)
cms.add('google.com')  # should return 1
cms.add('facebook.com', 25)  # insert 25 at once; should return 25

Import pyprobables and setup a Cuckoo Filter

from probables import CuckooFilter
cko = CuckooFilter(capacity=100, max_swaps=10)
cko.add('google.com')
cko.check('facebook.com')  # should return False
cko.check('google.com')  # should return True

Import pyprobables and setup a Quotient Filter

from probables import QuotientFilter
qf = QuotientFilter(quotient=24)
qf.add('google.com')
qf.check('facebook.com')  # should return False
qf.check('google.com')  # should return True

Supplying a pre-defined, alternative hashing strategies

from probables import BloomFilter
from probables.hashes import default_sha256
blm = BloomFilter(est_elements=1000, false_positive_rate=0.05,
                  hash_function=default_sha256)
blm.add('google.com')
blm.check('facebook.com')  # should return False
blm.check('google.com')  # should return True

Defining hashing function using the provided decorators

import mmh3  # murmur hash 3 implementation (pip install mmh3)
from probables.hashes import hash_with_depth_bytes
from probables import BloomFilter

@hash_with_depth_bytes
def my_hash(key, depth):
    return mmh3.hash_bytes(key, seed=depth)

blm = BloomFilter(est_elements=1000, false_positive_rate=0.05, hash_function=my_hash)

import hashlib
from probables.hashes import hash_with_depth_int
from probables.constants import UINT64_T_MAX
from probables import BloomFilter

@hash_with_depth_int
def my_hash(key, seed=0, encoding="utf-8"):
    max64mod = UINT64_T_MAX + 1
    val = int(hashlib.sha512(key.encode(encoding)).hexdigest(), 16)
    val += seed  # not a good example, but uses the seed value
    return val % max64mod

blm = BloomFilter(est_elements=1000, false_positive_rate=0.05, hash_function=my_hash)

See the API documentation for other data structures available and the quickstart page for more examples!

Changelog

Please see the changelog for a list of all changes.

Backward Compatible Changes

If you are using previously exported probablistic data structures (v0.4.1 or below) and used the default hashing strategy, you will want to use the following code to mimic the original default hashing algorithm.

from probables import BloomFilter
from probables.hashes import hash_with_depth_int

@hash_with_depth_int
def old_fnv1a(key, depth=1):
    return tmp_fnv_1a(key)

def tmp_fnv_1a(key):
    max64mod = UINT64_T_MAX + 1
    hval = 14695981039346656073
    fnv_64_prime = 1099511628211
    tmp = map(ord, key)
    for t_str in tmp:
        hval ^= t_str
        hval *= fnv_64_prime
        hval %= max64mod
    return hval

blm = BloomFilter(filpath="old-file-path.blm", hash_function=old_fnv1a)