kDCI: on using direct count up to the third iteration

Lucchese C, Orlando S, Perego R

Frequent itemsets mining 

We thus introduced such technique in the last version of kDCI, which is level-wise hybrid algorithm. kDCI stores the dataset with an horizontal format to disk during the first iterations. After some iteration the dataset may become small enough (thanks to anti-monotone frequency pruning) to be stored in the main memory in a vertical format, and after that the algorithm goes on performing tid-lists intersections to retrieve itemsets supports, and searches among candidates are not needed anymore. Usually the dataset happens to be small enough at most at the fourth iteration.

Publisher: CEUR-WS.org