Greedy Selection Clause Samples

Greedy Selection. ‌ ▇▇▇▇▇▇ et al. [43] compared the pattern selection method to the Gamer approach [82], which tries to construct one single best PDB for a problem. Its pattern selection method is an iterative process, starting with all the goal variables in one pattern, where the causally connected variables, who would most increase the average h value of the associated PDB, are added to the pattern. Following this work, we devised a new Gamer-style pattern generation method, which behaves similarly, but which adds the option of partial pattern database generations to it. By partial we mean that we have a time and memory limit for building each PDB. If the PDB building goes past this limit, we truncate it in the same way we would do with a perimeter PDB, i.e. any unmapped real state has the biggest h value when the PDB creation was interrupted. An important difference with the Gamer method is that we do not try every possible pattern resulting in an addition of a single causally connected variable to the latest pattern.