[...]... make arbitrarily small 4 ! for i ! 1 in typical scenarios: We envision as a typical scenario, that the user initially changes the support threshold often, in reaction to the continously generated association rules After the user has found a satisfactory threshold we do not expect further changes in the support threshold, i.e = = : : : for some transaction index j Hence ! for i ! 1 5 Achieving = for... Figure 6 We note that although we formally pass the support sequence as a parameter to PhaseI, is actually determined by the user while PhaseI scans the transaction sequence The generation of the association rules from the support lattice and the displaying of them to the user should also be placed in PhaseI and PhaseII An implementation of CARMA is under way In the following theorem we state the correctness