We present a novel algorithm to compute large itemsets online.
The user is free to change the support threshold
any time during the first scan of the transaction sequence.
The algorithm maintains a superset of all large itemsets and for each
itemset a shrinking, deterministic interval on its support.
After at most 2 scans the algorithm terminates with the precise support
for each large itemset.
Typically our algorithm is by an order of magnitude
more memory efficient than Apriori or DIC.