... set.¹

In our current implementation the result set is implemented as a bit-matrix indexed on queries and examples. This implementation is practically feasible (on typical computers at the time of writing) even when the number of queries in the pack multiplied by the number of examples is up to a billion, a bound which holds for most current ILP applications.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... data,²

I.e., that subset of the original data set for which the parent query succeeded; or, in the decision tree context: the examples sorted into the node that is being refined.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... refinements³

Refinements found to be specialisations of infrequent queries cannot be frequent themselves, and are pruned consequently.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... request.⁴

See http://www.cs.kuleuven.ac.be/~dtai/ACE/.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... obtained.⁵

In one case, with a relatively small pack, the system became slower. The timings indicate that this is not due to the compilation time, but to other changes in the implementation which for this relatively simple problem were not compensated by the faster execution of the packs.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.