apriori_analysis.py

# Marked Basked Optimisation using Apriori Rule Learning

# We are going to use this implementation of the `apriori` algorithm: https://github.com/ymoch/apyori
# For this case, the source code is directly included in the same folder as this file.

# Importing the libraries
import pandas as pd


# Import the dataset
data_path = 'annotated-code/Part 5 - Association Rule Learning/Section 28 - Apriori/Market_Basket_Optimisation.csv'
dataset_frame = pd.read_csv(data_path, header=None)  # The data has no headers
# The Apyori algorithm expects a list of lists, not a dataframe
transactions = []
for row in dataset_frame.values:
    transactions.append([item for item in row if not pd.isnull(item)])

# Dummy test data
# transactions = [
#     ['bread', 'milk'],
#     ['bread', 'diaper', 'beer', 'eggs'],
#     ['milk', 'diaper', 'beer', 'coke'],
#     ['milk', 'diaper', 'beer', 'bread'],
#     ['milk', 'diaper', 'bread', 'coke']
# ]

# Use the Apyori library
import sys # needed in this particular project to find the baked in apyori library
sys.path.insert(1, 'annotated-code/Part 5 - Association Rule Learning/Section 28 - Apriori')
from apyori import apriori
rules_generator = apriori(transactions, min_support=0.003, min_confidence=0.2, min_lift=3)
# Rules are lazily evaluated and generated here
# The rules returned by apyori are not ordered in any way
unordered_rules = list(rules_generator)


# Clean and sort the rules
# Apyori returns the rules in a non-intuitive and nested data structure, this flattens it up
cleaned_rules = []
for itemset in unordered_rules:
    for rule in itemset.ordered_statistics:
        cleaned_rules.append({
            'items_base': rule.items_base,
            'items_add': rule.items_add,
            'support': itemset.support,
            'confidence': rule.confidence,
            'lift': rule.lift
        })
from operator import itemgetter
cleaned_sorted_rules = sorted(cleaned_rules, key=itemgetter('lift'), reverse=True)

# Store them in a dataframe for easier visualization
rules_df = pd.DataFrame(cleaned_sorted_rules, columns=['items_base', 'items_add', 'support', 'confidence', 'lift'])