15.08.2013 Views

General Computer Science 320201 GenCS I & II Lecture ... - Kwarc

General Computer Science 320201 GenCS I & II Lecture ... - Kwarc

General Computer Science 320201 GenCS I & II Lecture ... - Kwarc

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

A ∗ search: Properties<br />

Complete Yes (unless there are infinitely many nodes n with f(n) ≤ f(0))<br />

Time Exponential in [relative error in h × length of solution]<br />

Space Same as time (variant of BFS)<br />

Optimal Yes<br />

A ∗ expands all (some/no) nodes with f(n) < h ∗ (n)<br />

The run-time depends on how good we approximated the real cost h ∗ with h.<br />

c○: Michael Kohlhase 488<br />

Since the availability of admissible heuristics is so important for informed search (particularly for<br />

A ∗ ), let us see how such heuristics can be obtained in practice. We will look at an example, and<br />

then derive a general procedure from that.<br />

Admissible heuristics: Example 8-puzzle<br />

Example 608 Let h1(n) be the number of misplaced tiles in node n (h1(S) = 6)<br />

Example 609 Let h2(n) be the total manhattan distance from desired location of each tile.<br />

(h2(S) = 2 + 0 + 3 + 1 + 0 + 1 + 3 + 4 = 14)<br />

Observation 610 (Typical search costs) (IDS ˆ= iterative deepening search)<br />

Dominance<br />

nodes explored IDS A ∗ (h1) A ∗ (h2)<br />

d = 14 3,473,941 539 113<br />

d = 24 too many 39,135 1,641<br />

c○: Michael Kohlhase 489<br />

Definition 611 Let h1 and h2 be two admissible heuristics we say that h2 dominates h1 if<br />

h2(n) ≥ h1(n) or all n.<br />

Theorem 612 If h2 dominates h1, then h2 is better for search than h1.<br />

Relaxed problems<br />

c○: Michael Kohlhase 490<br />

Finding good admissible heuristics is an art!<br />

Idea: Admissible heuristics can be derived from the exact solution cost of a relaxed version<br />

of the problem.<br />

258

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!