Sampling
Use a sampling process to generate candidate sets
Randomly select subset of transactions to consider
Negative Border, Bd-(S) : Set of minimal itemset sets X where X is not in S
Scan D using S union Bd-(S) as candidate set
Expand by adding negative border recursively until negative border empty
Worse case - 2 passes, Best case - 1 pass
First scan may be very inefficient
Small non-zero probability of missing itemset, for complete accuracy use entire database
With Partitioning, may view 1st partition as sample