Efficient-Apriori

An efficient pure Python implementation of the Apriori algorithm. Works with Python 3.7+.

The apriori algorithm uncovers hidden structures in categorical data. The classical example is a database containing purchases from a supermarket. Every purchase has a number of items associated with it. We would like to uncover association rules such as {bread, eggs} -> {bacon} from the data. This is the goal of association rule learning, and the Apriori algorithm is arguably the most famous algorithm for this problem. This repository contains an efficient, well-tested implementation of the apriori algorithm as described in the original paper by Agrawal et al, published in 1994.

The code is stable and in widespread use. It's cited in the book "Mastering Machine Learning Algorithms" by Bonaccorso.

The code is fast. See timings in this PR.

Example

Here's a minimal working example. Notice that in every transaction with eggs present, bacon is present too. Therefore, the rule {eggs} -> {bacon} is returned with 100 % confidence.

from efficient_apriori import apriori
transactions = [('eggs', 'bacon', 'soup'),
                ('eggs', 'bacon', 'apple'),
                ('soup', 'bacon', 'banana')]
itemsets, rules = apriori(transactions, min_support=0.5, min_confidence=1)
print(rules)  # [{eggs} -> {bacon}, {soup} -> {bacon}]

If your data is in a pandas DataFrame, you must convert it to a list of tuples. Do you have missing values, or does the algorithm run for a long time? See this comment. More examples are included below.

Installation

The software is available through GitHub, and through PyPI. You may install the software using pip.

pip install efficient-apriori

Contributing

You are very welcome to scrutinize the code and make pull requests if you have suggestions and improvements. Your submitted code must be PEP8 compliant, and all tests must pass. See list of contributors here.

More examples

Filtering and sorting association rules

It's possible to filter and sort the returned list of association rules.

from efficient_apriori import apriori
transactions = [('eggs', 'bacon', 'soup'),
                ('eggs', 'bacon', 'apple'),
                ('soup', 'bacon', 'banana')]
itemsets, rules = apriori(transactions, min_support=0.2, min_confidence=1)

# Print out every rule with 2 items on the left hand side,
# 1 item on the right hand side, sorted by lift
rules_rhs = filter(lambda rule: len(rule.lhs) == 2 and len(rule.rhs) == 1, rules)
for rule in sorted(rules_rhs, key=lambda rule: rule.lift):
  print(rule)  # Prints the rule and its confidence, support, lift, ...

Transactions with IDs

If you need to know which transactions occurred in the frequent itemsets, set the output_transaction_ids parameter to True. This changes the output to contain ItemsetCount objects for each itemset. The objects have a members property containing is the set of ids of frequent transactions as well as a count property. The ids are the enumeration of the transactions in the order they appear.

from efficient_apriori import apriori
transactions = [('eggs', 'bacon', 'soup'),
                ('eggs', 'bacon', 'apple'),
                ('soup', 'bacon', 'banana')]
itemsets, rules = apriori(transactions, output_transaction_ids=True)
print(itemsets)
# {1: {('bacon',): ItemsetCount(itemset_count=3, members={0, 1, 2}), ...