Available algorithms

Candidate generation based

ModalAssociationRules.apriori — Function

apriori(miner::Miner; verbose::Bool=true)::Nothing

Apriori algorithm, as described here but generalized to also work with modal logic.

Arguments

miner::M: miner containing the data and the extraction parameterization;
prune_strategy::Function=grow_prune: strategy to prune candidates between one iteration

and the successive;

verbose::Bool=false: print informations about each iteration.

source

TreeProjection based

FPGrowth

ModalAssociationRules.fpgrowth — Function

fpgrowth(miner::AbstractMiner; verbose::Bool=true)::Nothing

(Modal) FP-Growth algorithm, as described here.

Arguments

miner: miner containing the extraction parameterization;
X: data from which you want to mine association rules;
parallel: enable multi-threaded execution, using Threads.nthreads() threads;
distributed: enable multi-processing execution, with Distributed.nworkers() processes;
verbose: print detailed informations while the algorithm runs.

Requirements

This implementation requires a custom Bulldozer constructor capable of handling the given AbstractMiner. In particular, the following dispatch must be implemented:

Bulldozer(miner::MyMinerType, ith_instance::Integer)

source

FPGrowth algorithm relies on two data structures, FPTree and HeaderTable. To know more about them and their, please refer to the documentation here data-structures.

FPGrowth algorithm relies on the following two routines.

ModalAssociationRules.patternbase — Method

patternbase(
    item::Item,
    htable::HeaderTable,
    miner::AbstractMiner
)::ConditionalPatternBase

Retrieve the ConditionalPatternBase of fptree based on item.

The conditional pattern based on a FPTree is the set of all the paths from the tree root to nodes containing item (not included). Each of these paths is represented by an EnhancedItemset.

The EnhancedItemsets in the returned ConditionalPatternBase are sorted decreasingly by gsupport.

source

ModalAssociationRules.projection — Method

function projection(pbase::ConditionalPatternBase, miner::AbstractMiner)

Return respectively a FPTree and a HeaderTable starting from pbase. An AbstractMiner must be provided to guarantee the generated header table internal state is OK, that is, its items are sorted decreasingly by gsupport.

source

Also, FPGrowth requires the Miner to remember the worlds associated with the extracted frequent itemsets. To add this functionality, we can define a new dispatch of initminingstate: it is automatically considered to enrich the miner, while building it together with fpgrowth as mining algorithm.

ModalAssociationRules.initminingstate — Method

initminingstate(::typeof(fpgrowth), ::MineableData)::MiningState

MiningState fields levereged when executing FP-Growth algorithm.

source

Anchored semantics

ModalAssociationRules.anchored_semantics — Function

anchored_semantics(miner::M; kwargs...)::M where {M<:AbstractMiner}

Logic to be executed before the the algorithm wrapped within the miner; the goal is to make such an algorithm coherent with anchored semantics.

source

ModalAssociationRules.anchored_apriori — Function

anchored_apriori(miner::AbstractMiner; kwargs...)::Nothing

Anchored version of apriori algorithm, that is exactly apriori but assuring that miner possess atleast isanchored_itemset policy, with ignoreuntillength parameter set to 1 or higher.

TODO - insert a reference to TIME2025 article.

source

ModalAssociationRules.anchored_fpgrowth — Function

function anchored_fpgrowth(miner::M; kwargs...)::M where {M<:AbstractMiner}

Implementation of fpgrowth with anchored semantics. Essentially, Items are SoleData.VariableDistances wrapping motifs.

More information about motifs: <insert-link> More information about the implementation: <insert-link>

source