Cuckoo Filter implementation in Python
Cuckoo Filter, like Bloom Filter, is a probabilistic data structure for fast, approximate set membership queries, with some small false positive probability. While Bloom Filters are space efficient and are widely used, they do not support deletion of items from the set without rebuilding the entire filter. This can be overcome with several extensions to Bloom Filters such as Counting Bloom Filters, but with significant space overhead.
Cuckoo Filters support adding and removing items dynamically while achieving higher performance than Bloom filters. A Cuckoo Filter is based on partial-key cuckoo hashing that stores only fingerprint of each item inserted. Cuckoo Filters provide higher lookup performance than Bloom Filters and uses less space than Bloom Filters if the target false positive rate is < 3%.
The original research paper Cuckoo Filter: Practically Better Than Bloom by Bin Fan, David G. Andersen, Michael Kaminsky and Michael D. Mitzenmacher describes the data structure in more detail.
Install cuckoopy using:
$ pip install cuckoopy
>>> from cuckoopy import CuckooFilter # Initialize a cuckoo filter with 10000 buckets with bucket size 4 and fingerprint size of 1 byte >>> cf = CuckooFilter(capacity=10000, bucket_size=4, fingerprint_size=1)
Insert an item into the filter:
>>> cf.insert('Hello!') True
Lookup an item in the filter:
>>> cf.contains('Hello!') True >>> 'Hello!' in cf True
Delete an item from the filter:
>>> cf.delete('Hello!') True
Get the size (number of items present) of the filter:
>>> cf.size 4 >>> len(cf) 4
Running tests locally
This project uses pytest for tests. Make sure you have tox installed on your local machine and from the root directory of the project, run:
This command runs unit tests in python 3.5 and python 3.6 environments with code coverage details. It also runs pep8 (flake8) checks. To run tox against a specific environment (py35, py36 or pep8), use the -e option.