Due to the incremental nature of PDL2, and to limit the effects of ordering in learning, quantities are also updated for the training pair. This is reflected in the symmetry of the algorithm (e.g., subset and superset cases are reciprocal).
Due to the incremental nature of PDL2, and tolimit the effects of ordering in learning, quantities arealso updated for the training pair. This is reflected inthe symmetry of the algorithm (e.g., subset andsuperset cases are reciprocal).