Mercury Settrie

Python 3.8 Python 3.9 Python 3.10 Python 3.11 Apache 2 license Ask Me Anything !

What is this?

TLDR: A SetTrie is a container of sets that performs efficient subset and superset queries.

Settrie is a Python library implemented in C++ to create, update and query SetTrie objects.

What problems is Settrie good for?

Settrie was born from the need of a better implementation of the algorithm for our recommender system. It has direct application to text indexing for search in logarithmic time of documents containing a set of words. Think of a collection of documents as a container of the set of words that each document has. Finding all the documents that contain a set of words is finding the superset of the set of words in the query. A SetTrie will give you the answer –the list of document names– in logarithmic time. The data structure is also used in auto-complete applications.

What data size can Settrie tackle?

Settrie is a C++ implementation with a Python interface. It is single threaded and can seamlessly operate over really large collections of sets. Note that the main structure is a tree and a tree node is exactly 20 bytes, a billion nodes is 20 Gb, of course plus some structures to store identifiers, etc. Note that the tree is compressing the documents by sharing the common parts and documents are already compressed by considering them a set of words. An of-the-shelf computer can store in RAM a representation of terabytes of documents and query result in much less than typing speed.

How does this implementation compare to other Python implementations?

It is about 200 times faster and 20 times more memory efficient that a pure Python implementation.

Try it without any installation in Google Colab

The API is very easy to use. You can see this benchmark notebook for reference.


pip install mercury-settrie


from settrie import SetTrie

# Create a SetTrie object
stt = SetTrie()

# Insert some sets
stt.insert({2, 3}, 'id1')
stt.insert({2, 3, 4.4}, 'id2')
stt.insert({'Mon', 'Tue'}, 'days')

# Find id by set
print(stt.find({2, 3}))

# Find ids of all supersets
for id in stt.supersets({2, 3}):

# Find ids of all subsets
for id in stt.subsets({2, 3}):

# Nested iteration over the sets and elements
for st in stt:
    for e in st.elements:
        print('  ', e)

# Store as a pickle file file
import pickle
with open('my_settrie.pickle', 'wb') as f:
    pickle.dump(stt, f)

# Load from a pickle file
with open('my_settrie.pickle', 'rb') as f:
    tt = pickle.load(f)

# Check that they are identical
for t, st in zip(tt, stt):
    assert ==
    for et, est in zip(t.elements, st.elements):
        assert et == est

# Remove sets by id

# After many .remove() calls, the tree has nodes marked as dirty,
# calling .purge() removes them completely and frees RAM.

How to setup a development environment and contribute to settrie


Thank you to our contributor(s)



                         Apache License
                   Version 2.0, January 2004

     Copyright 2021-23, Banco de Bilbao Vizcaya Argentaria, S.A.

   Licensed under the Apache License, Version 2.0 (the "License");
   you may not use this file except in compliance with the License.
   You may obtain a copy of the License at