guenthermi / fast_minh

Python package for fast MinHash calculation and operations

fast_minh

A small Python package for calculating MinHash values, computing approaximated Jaccard similarity, and building LSH indices of MinHash values to perform fast approximated set similarity search.

Installation

pip install .

Usage

Calculating MinHash values:

To calculate min hash values, you can create a HashFamily object which initializes a set of hash fuctions (default: 128). After that you can use the HashFamily.minh function to obtain a set of MinHash values for a given set of strings:

hf = HashFamily()
mh = hf.minh(['test', 'it', 'out'])

Calculate an approximated Jaccard coefficient:

After calculating multiple minhash values for different sets with the same hash family, you can use the jaccard function to determine and approximated similarity score:

from fast_minh import minh, jaccard
hf = HashFamily()
mh1 = hf.minh(['test', 'it', 'out'])
mh2 = hf.minh(['test', 'it', 'again'])
sim = jaccard(mh1, mh2)

MinHash LSH Index:

To find similar sets of text values fast, you can use an MinHash LSH index. You can insert sets with the LshIndex.insert function and retrieve similar candidates with the LshIndex.find method:

from fast_minh import LshIndex
lsh = LshIndex(1, 3)
input_key = 'Key'
input_set = ['A', 'set', 'of', 'multiple', 'tokens']
lsh.insert(input_key, input_set)
out = lsh.find(input_set)