Self-tuning miner of filtered association rules.
Computes closure-based association rules using confidence boost as a criterion to suppress intuitively redundant rules. This node is intended to be usable with no parameter tuning, although it does support some advanced configuration.
Due to inherent complexities of computation, this node may take several minutes to complete. Be patient.
For dense datasets, you might need to increase the available memory for the JVM; to do this, tweak the -Xmx parameter in file knime.ini (read further on in the KNIME documentation if necessary).
The best way to use this node is along with a File Reader node to read the transactions as strings: set delimiter (often a space) and check "allow short lines" permission in the "advanced" area. Connect it to a Create Collection Column node - preferably with "set" option checked in case items might be repeated on transactions. "Missing values" are automatically ignored. The resulting output is a valid input for this node.
The first output port of the node contains rules given as antecedent and consequent collections along with its details. The second output has gone through some string formatting and is appropriate to be written to a file (e.g. via a CSV Writer node).
This node has been implemented in cooperation with Universidad de Cantabria and Universitat Politecnica de Catalunya with partial support from Pascal 2 through the STARK project of the Harvest Programme .