FIM.Eclat -- DECLAT / FuzzyDECLAT (Spark)

Requires the spark extra (pip install -e ".[spark]", see Installation).

Frequent itemset mining on Apache Spark inspired by the sequential ECLAT algorithm (Zaki, 2000), which represents each item by its list of transaction identifiers (TID-list) and computes an itemset's support by intersecting the TID-lists of its items.

DECLAT (crisp)

Implements Algorithm 3 of Fernandez-Basso, Ruiz & Martin-Bautista (2024) -- see Citing ARMxtend.

DECLAT.run(sc, transactions, min_supp)

Same arguments and return type as FIM.apriori.DApriori (RDD[str] transactions, ,-separated items, dict {itemset_key: support} result) -- the three algorithms (DApriori/DAprioriTID/DECLAT) are interchangeable and produce identical results, they only differ in how the counting is distributed internally.

from ARMxtend.FIM.Eclat import DECLAT

freq_itemsets = DECLAT.run(sc, transactions, min_supp=0.5)

FuzzyDECLAT (fuzzy)

The fuzzy, alpha-cut-based counterpart, consistent with FIM.BD_FARE.FuzzyDAprioriTID and FFIM.fuzzy_fpgrowth (same input convention, same output format, interchangeable as input to FIM.FARE.fuzzy_association_rules).

FuzzyDECLAT.run(sc, transactions, min_supp, num_alpha)

• transactions (RDD[Iterable[Tuple[str, float]]]): each transaction is a list of (item, membership degree in [0, 1]) pairs.
• num_alpha (int): number of alpha-cuts (10 is a reasonable default).

Returns dict {itemset_key: numpy.ndarray(num_alpha)}: the bit-list of relative support of each frequent itemset at every alpha-cut. An itemset is frequent if its aggregated fuzzy support FSupp (see FIM.FARE) reaches min_supp.

from ARMxtend.FIM.Eclat import FuzzyDECLAT

fuzzy_transactions = sc.parallelize([
    [("cold", 1.0), ("low_humidity", 0.8)],
    [("cold", 0.9), ("low_humidity", 0.6)],
    [("warm", 0.7), ("low_humidity", 0.9)],
])
freq_itemsets = FuzzyDECLAT.run(sc, fuzzy_transactions, min_supp=0.3, num_alpha=10)