Searching Artificial Intelligence Spring 2009 Problemsolving agents Example
Searching Artificial Intelligence Spring 2009
Problem-solving agents
Example: Romania n On holiday in Romania; currently in Arad. Flight leaves tomorrow from Bucharest n Formulate goal: n ¨ be n in Bucharest Formulate problem: ¨ states: various cities ¨ actions: drive between cities n Find solution:
Example: Romania
Problem types n Deterministic, fully observable single-state problem ¨ n Non-observable sensorless problem (conformant problem) ¨ n Agent knows exactly which state it will be in; solution is a sequence Agent may have no idea where it is; solution is a sequence Nondeterministic and/or partially observable contingency problem percepts provide new information about current state ¨ often interleave} search, execution ¨ n Unknown state space exploration problem
Example: vacuum world n Single-state, start in #5. Solution?
Example: vacuum world n Single-state, start in #5. Solution? [Right, Suck] n Sensorless, start in {1, 2, 3, 4, 5, 6, 7, 8} e. g. , Right goes to {2, 4, 6, 8} Solution?
Example: vacuum world n Sensorless, start in {1, 2, 3, 4, 5, 6, 7, 8} e. g. , Right goes to {2, 4, 6, 8} Solution? [Right, Suck, Left, Suck] n Contingency Nondeterministic: Suck may dirty a clean carpet ¨ Partially observable: location, dirt at current location. ¨ Percept: [L, Clean], i. e. , start in #5 or #7 ¨ Solution?
Single-state problem formulation A problem is defined by four items: 2. initial state e. g. , "at Arad" actions or successor function S(x) = set of action–state pairs ¨ e. g. , S(Arad) = {<Arad Zerind, Zerind>, … } 3. goal test, can be 1. ¨ ¨ 4. path cost (additive) ¨ ¨ n explicit, e. g. , x = "at Bucharest" implicit, e. g. , Checkmate(x) e. g. , sum of distances, number of actions executed, etc. c(x, a, y) is the step cost, assumed to be ≥ 0 A solution is a sequence of actions leading from the initial state to a goal state
Selecting a state space n Real world is absurdly complex state space must be abstracted for problem solving n n (Abstract) state set of real states (Abstract) action complex combination of real actions ¨ e. g. , "Arad Zerind" represents a complex set of possible routes, detours, rest stops, etc. n For guaranteed realizability, any real state "in Arad“ must get to some real state "in Zerind" n (Abstract) solution = ¨ n set of real paths that are solutions in the real world Each abstract action should be "easier" than the original problem
Vacuum world state space graph n n states? actions? goal test? path cost?
Vacuum world state space graph n n states? integer dirt and robot location actions? Left, Right, Suck goal test? no dirt at all locations path cost? 1 per action
Example: The 8 -puzzle n n states? actions? goal test? path cost?
Example: The 8 -puzzle n n states? locations of tiles actions? move blank left, right, up, down goal test? = goal state (given) path cost? 1 per move [Note: optimal solution of n-Puzzle family is NP-hard]
Example: robotic assembly n n states? : real-valued coordinates of robot joint angles parts of the object to be assembled actions? : continuous motions of robot joints goal test? : complete assembly path cost? : time to execute
Tree search algorithms n Basic idea: ¨ offline, simulated exploration of state space by generating successors of already-explored states (a. k. a. ~expanding states)
Tree search example
Tree search example
Tree search example
Implementation: general tree search
Implementation: states vs. nodes n n n A state is a (representation of) a physical configuration A node is a data structure constituting part of a search tree includes state, parent node, action, path cost g(x), depth The Expand function creates new nodes, filling in the various fields and using the Successor. Fn of the problem to create the corresponding states.
Search strategies n n A search strategy is defined by picking the order of node expansion Strategies are evaluated along the following dimensions: completeness: does it always find a solution if one exists? ¨ time complexity: number of nodes generated ¨ space complexity: maximum number of nodes in memory ¨ optimality: does it always find a least-cost solution? ¨ n Time and space complexity are measured in terms of b: maximum branching factor of the search tree ¨ d: depth of the least-cost solution ¨ m: maximum depth of the state space (may be ∞) ¨
Uninformed search strategies use only the information available in the problem definition n Breadth-first search n Uniform-cost search n Depth-first search n Depth-limited search n Iterative deepening search n
Breadth-first search Expand shallowest unexpanded node n Implementation: n ¨ fringe is a FIFO queue, i. e. , new successors go at end A
Breadth-first search Expand shallowest unexpanded node n Implementation: n ¨ fringe is a FIFO queue, i. e. , new successors go at end A B C
Breadth-first search n Expand shallowest unexpanded node A B C D E
Breadth-first search n Expand shallowest unexpanded node A B C D E F G
Properties of breadth-first search n Complete? Yes (if b is finite) Time? 1+b+b 2+b 3+… +bd + b(bd-1) = O(bd+1) Space? O(bd+1) (keeps every node in memory) Optimal? Yes (if cost = 1 per step) n Space is the bigger problem (more than time) n n n
Uniform-cost search n n Expand least-cost unexpanded node Implementation: fringe = priority queue ¨ Equivalent to breadth-first if step costs all equal ¨ n n Complete? Yes, if step cost ≥ ε Time? # of nodes with g ≤ cost of optimal solution, O(bceiling(C*/ ε)) where C* is the cost of the optimal solution Space? # of nodes with g ≤ cost of optimal solution, O(bceiling(C*/ ε)) Optimal? Yes – nodes expanded in increasing order of g(n)
Depth-first search n n Expand deepest unexpanded node Implementation: ¨ fringe = Stack, i. e. , put successors at front A push A;
Depth-first search n n Expand deepest unexpanded node Implementation: ¨ fringe = Stack B A C pop A; push C; push B;
Depth-first search n n Expand deepest unexpanded node Implementation: B ¨ fringe = STACK D C E C pop B; push E; push D;
Depth-first search n Expand deepest unexpanded node D H E I C E pop D; push I; push H; C
Depth-first search n Expand deepest unexpanded node H I E C I E pop H; C
Depth-first search n Expand deepest unexpanded node I E C pop I;
Depth-first search n Expand deepest unexpanded node E C J K pop E; push K; push J C
Depth-first search n Expand deepest unexpanded node J K K C C pop J;
Depth-first search n Expand deepest unexpanded node K C C pop K;
Depth-first search n Expand deepest unexpanded node C F G pop C; push G; push F;
Depth-first search n Expand deepest unexpanded node F L G M pop F; push M; push L; G
Depth-first search n Expand deepest unexpanded node L M M G G pop L;
Properties of depth-first search n Complete? No: fails in infinite-depth spaces, spaces with loops ¨ Modify to avoid repeated states along path complete in finite spaces n Time? O(bm): terrible if m is much larger than d ¨ n n but if solutions are dense, may be much faster than breadth-first Space? O(bm), i. e. , linear space! Optimal? No
Depth-limited search = depth-first search with depth limit l, i. e. , nodes at depth l have no successors n Recursive implementation:
Iterative deepening search
Iterative deepening search l =0
Iterative deepening search l =1
Iterative deepening search l =2
Iterative deepening search l =3
Iterative deepening search n Number of nodes generated in a depth-limited search to depth d with branching factor b: NDLS = b 0 + b 1 + b 2 + … + bd-2 + bd-1 + bd n Number of nodes generated in an iterative deepening search to depth d with branching factor b: NIDS = (d+1)b 0 + d b^1 + (d-1)b^2 + … + 3 bd-2 +2 bd-1 + 1 bd n For b = 10, d = 5: ¨ ¨ n NDLS = 1 + 100 + 1, 000 + 100, 000 = 111, 111 NIDS = 6 + 50 + 400 + 3, 000 + 20, 000 + 100, 000 = 123, 456 Overhead = (123, 456 - 111, 111)/111, 111 = 11%
Properties of iterative deepening search Complete? Yes n Time? (d+1)b 0 + d b 1 + (d-1)b 2 + … + bd = O(bd) n Space? O(bd) n Optimal? Yes, if step cost = 1 n
Summary of algorithms
Repeated states n Failure to detect repeated states can turn a linear problem into an exponential one!
Graph search
Summary n Problem formulation usually requires abstracting away real-world details to define a state space that can feasibly be explored n Variety of uninformed search strategies n Iterative deepening search uses only linear space and not much more time than other uninformed algorithms
Informed search algorithms Artificial Intelligence
Review: Tree search An abstract problem is modeled as a (finite or infinite) decision tree. n Weak methods do not use any knowledge of the problem n ¨ General applicable ¨ Usually die from combinatorial explosion when exposed to “real life”
Best-first search n Idea: use an evaluation function f(n) for each node estimate of "desirability" Expand most desirable unexpanded node ¨ n Implementation: Order the nodes in fringe in decreasing order of desirability n Special cases: greedy best-first search ¨ A* search ¨
Romania with step costs in km
Greedy best-first search n Evaluation function f(n) = h(n) (heuristic) ¨= estimate of cost from n to goal e. g. , h. SLD(n) = straight-line distance from n to Bucharest n Greedy best-first search expands the node that appears to be closest to goal n
Greedy best-first search example
Greedy best-first search example
Greedy best-first search example
Greedy best-first search example
Properties of greedy best-first search Complete? No – can get stuck in loops, e. g. , Iasi Neamt … n Time? O(bm), but a good heuristic can give dramatic improvement n Space? O(bm) -- keeps all nodes in memory n Optimal? No n
A* search Idea: avoid expanding paths that are already expensive n Evaluation function f(n) = g(n) + h(n) n ¨ g(n) = cost so far to reach n ¨ h(n) = estimated cost from n to goal n f(n) = estimated total cost of path through n to goal
A* search example
A* search example
A* search example
A* search example
A* search example
A* search example
Admissible heuristics n n A heuristic h(n) is admissible if for every node n, h(n) ≤ h*(n), where h*(n) is the true cost to reach the goal state from n. An admissible heuristic never overestimates the cost to reach the goal, i. e. , it is optimistic Example: h. SLD(n) (never overestimates the actual road distance) Theorem: If h(n) is admissible, A* using TREESEARCH is optimal
Optimality of A* (proof) n Suppose some suboptimal goal G 2 has been generated and is in the fringe. Let n be an unexpanded node in the fringe such that n is on a shortest path to an optimal goal G. n f(G 2) = g(G 2) > g(G) f(G) = g(G) f(G 2) > f(G) n n n since h(G 2) = 0 since G 2 is suboptimal since h(G) = 0 from above
Optimality of A* (proof) n Suppose some suboptimal goal G 2 has been generated and is in the fringe. Let n be an unexpanded node in the fringe such that n is on a shortest path to an optimal goal G. n f(G 2) h(n) g(n) + h(n) f(n) n n n > f(G) ≤ h^*(n) ≤ g(n) + h*(n) ≤ f(G) * from above since h is admissible
Consistent heuristics n A heuristic is consistent if for every node n, every successor n' of n generated by any action a, n h(n) ≤ c(n, a, n') + h(n') n If h is consistent, we have f(n') = g(n') + h(n') = g(n) + c(n, a, n') + h(n') ≥ g(n) + h(n) = f(n) n i. e. , f(n) is non-decreasing along any path.
Optimality of A* n A* expands nodes in order of increasing f value Gradually adds "f-contours" of nodes ¨ Contour i has all nodes with f=fi, where fi < fi+1 ¨
Properties of A* Complete? Yes (unless there are infinitely many nodes with f ≤ f(G) ) n Time? Exponential n Space? Keeps all nodes in memory n Optimal? Yes n
Admissible heuristics E. g. , for the 8 -puzzle: h 1(n) = number of misplaced tiles n h 2(n) = total Manhattan distance (i. e. , no. of squares from desired location of each tile) n n n h 1(S) = ? h (S) = ?
Admissible heuristics E. g. , for the 8 -puzzle: h 1(n) = number of misplaced tiles n h 2(n) = total Manhattan distance (i. e. , no. of squares from desired location of each tile) n n n h 1(S) = ? 8 h (S) = ? 3+1+2+2+2+3+3+2 = 18
Dominance n n n If h 2(n) ≥ h 1(n) for all n (both admissible) then h 2 dominates h 1 h 2 is better for search Typical search costs (average number of nodes expanded): d=12 IDS = 3, 644, 035 nodes A*(h 1) = 227 nodes A*(h 2) = 73 nodes d=24 IDS = too many nodes A*(h 1) = 39, 135 nodes A*(h 2) = 1, 641 nodes
Relaxed problems n n A problem with fewer restrictions on the actions is called a relaxed problem The cost of an optimal solution to a relaxed problem is an admissible heuristic for the original problem If the rules of the 8 -puzzle are relaxed so that a tile can move anywhere, then h 1(n) gives the shortest solution If the rules are relaxed so that a tile can move to any adjacent square, then h 2(n) gives the shortest solution
Local search algorithms n In many optimization problems, the path to the goal is irrelevant; the goal state itself is the solution ¨ State space = set of "complete" configurations ¨ Find configuration satisfying constraints, e. g. , n- queens n In such cases, we can use local search algorithms ¨ keep a single "current" state, tries to improve it
Example: n-queens n Put n queens on an n × n board with no two queens on the same row, column, or diagonal
Hill-climbing search n "Like climbing Everest in thick fog with amnesia"
Hill-climbing search n Problem: depending on initial state, can get stuck in local maxima
Hill-climbing search: 8 -queens problem n n h = number of pairs of queens that are attacking each other, either directly or indirectly h = 17 for the above state
Hill-climbing search: 8 -queens problem A local minimum with h = 1
Simulated annealing search n Idea: escape local maxima by allowing some "bad" moves but gradually decrease their frequency
Properties of simulated annealing search n One can prove: If T decreases slowly enough, then simulated annealing search will find a global optimum with probability approaching 1 n Widely used in VLSI layout, airline scheduling, etc
Local beam search n Keep track of k states rather than just one ¨ Start ¨ At with k randomly generated states each iteration, all the successors of all k states are generated ¨ If any one is a goal state, stop; else select the k best successors from the complete list and repeat.
Genetic algorithms n n n A successor state is generated by combining two parent states Start with k randomly generated states (population) A state is represented as a string over a finite alphabet (often a string of 0 s and 1 s) Evaluation function (fitness function). Higher values for better states. Produce the next generation of states by selection, crossover, and mutation
Genetic algorithms n n n Fitness function: number of non-attacking pairs of queens (min = 0, max = 8 × 7/2 = 28) 24/(24+23+20+11) = 31% 23/(24+23+20+11) = 29% etc
Genetic algorithms
- Slides: 93