SP10 cs188 lecture 4 -- CSPs (2PP)

# SP10 cs188 lecture 4 -- CSPs (2PP) - CS 188 Artificial...

This preview shows pages 1–8. Sign up to view the full content.

1 CS 188: Artificial Intelligence Spring 2010 Lecture 4: A* wrap-up + Constraint Satisfaction 1/28/2010 Pieter Abbeel – UC Berkeley Many slides from Dan Klein Announcements s Project 0 (Python tutorial) is due today s If you don’t have a class account yet, pick one up after lecture s Written 1 (Search) is due today s Project 1 (Search) is out and due next week Thursday s Section/Lecture

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
2 Recap: Search s Search problem: s States (configurations of the world) s Successor function: a function from states to lists of (state, action, cost) triples; drawn as a graph s Start state and goal test General Tree Search s Important ideas: s Fringe s Expansion s Exploration strategy s Main question: which fringe nodes to explore? Detailed pseudocode is in the book!
3 A* Review s A* uses both backward costs g and forward estimate h: f(n) = g(n) + h(n) s A* tree search is optimal with admissible heuristics (optimistic future cost estimates) s Heuristic design is key: relaxed problems can help Admissible Heuristics s A heuristic h is admissible (optimistic) if: where is the true cost to a nearest goal s Example: s Coming up with admissible heuristics is most of what’s involved in using A* in practice. 15

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
4 Optimality of A*: Blocking Notation: s g(n) = cost to node n s h(n) = estimated cost from n to the nearest goal (heuristic) s f(n) = g(n) + h(n) = estimated total cost via n s G*: a lowest cost goal node s G: another goal node Optimality of A*: Blocking Proof: s What could go wrong? s We’d have to have to pop a suboptimal goal G off the fringe before G* s This can’t happen: s Imagine a suboptimal goal G is on the queue s Some node n which is a subpath of G* must also be on the fringe (why?) s n will be popped before G
5 Tree Search: Extra Work! s Failure to detect repeated states can cause exponentially more work. Why? Graph Search s Very simple fix: never expand a state twice s Can this wreck completeness? Optimality?

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
6 Optimality of A* Graph Search Proof: s New possible problem: nodes on path to G* that would have been in queue aren’t, because some worse n’ for the same state as some n was dequeued and expanded first (disaster!) s Take the highest such n in tree s Let p be the ancestor which was on the queue when n ’ was expanded s Assume f(p) < f(n) s f(n) < f(n’) because n’ is suboptimal s p would have been expanded before n s Contradiction! Consistency s Wait, how do we know parents have better f-values than their successors? s Couldn’t we pop some node n , and find its child n’ to have lower f value? s YES: s What can we require to prevent these inversions? s Consistency: s Real cost must always exceed reduction in heuristic A B G 3 h = 0 h = 10 g = 10
7 A* Graph Search Gone Wrong S A B C G 1 1 1 2 3 h=2 h=1 h=4 h=1 h=0 S (0+ 2 ) A (1+ 4 ) B (1+ 1 ) C (2+ 1 ) C (3+ 1 ) G (6+ 0 ) State space graph Search tree C is already in the closed-list, hence not placed in the priority queue Consistency 3 A C G h=4 h=1 1 The story on Consistency:

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

## This note was uploaded on 03/01/2010 for the course COMPUTER S 188 taught by Professor Abbel during the Spring '10 term at Berkeley.

### Page1 / 30

SP10 cs188 lecture 4 -- CSPs (2PP) - CS 188 Artificial...

This preview shows document pages 1 - 8. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online