Winnowing

A Python implementation of the Winnowing (local algorithms for document fingerprinting)

Original Work

The original research paper can be found at http://dl.acm.org/citation.cfm?id=872770.

Installation

You may install winnowing package via pip as follows:

pip install winnowing

Alternatively, you may also install the package by cloning this repository.

git clone https://github.com/suminb/winnowing.git
cd winnowing && python setup.py install

Usage

>>> from winnowing import winnow

>>> winnow('A do run run run, a do run run')
set([(5, 23942), (14, 2887), (2, 1966), (9, 23942), (20, 1966)])

>>> winnow('run run')
set([(0, 23942)]) # match found!

Default Hash Function

Quite honestly, I did not know what hash function to use. The paper did not talk about it. So I decided to use a part of SHA-1; more precisely, the last 16 bits of the digest.

Custom Hash Function

You may use your own hash function as demonstrated below.

def hash_md5(text):
    import hashlib

    hs = hashlib.md5(text)
    hs = hs.hexdigest()
    hs = int(hs, 16)

    return hs

# Override the hash function
winnow.hash_function = hash_md5

winnow('The cake was a lie')

Lower Bound of Fingerprint Density

(TODO: Write this section)