400 likes | 570 Views
King Saud University College of Computer and Information Sciences Information Technology Department IT422 - Intelligent systems . Chapter 3. PROBLEM SOLVING BY SEARCHING (2). Informed Search.
E N D
King Saud University College of Computer and Information Sciences Information Technology Department IT422 - Intelligent systems Chapter 3 PROBLEM SOLVING BY SEARCHING (2)
Informed Search • One that uses problem specific knowledge beyond the definition of the problem itself to guide the search. • Why? • Without incorporating knowledge into searching, one is forced to look everywhere to find the answer. Hence, the complexity of uninformed search is intractable. • With knowledge, one can search the state space as if he was given “hints” when exploring a maze. • Heuristic information in search = Hints • Leads to dramatic speed up in efficiency.
G L K J I H F N E C B A M D O Informed Search • Best-First Search • Greedy Best First Search • A* Search • Local search algorithms • Stochastic Search algorithms Search only in this subtree!!
Best first search • Key idea: • Use an evaluation function f(n) for each node: • estimate of “distance” to the goal. • Node with the lowest evaluation is chosen for expansion. • Implementation: • frontier: maintain the frontier in ascending order of f-values • Special cases: • Greedy best-first search • A* search
Formal description of Best-First Search algorithm Function Best-First Graph-Search(problem,frontier,f) returns a solution or a failure // f: evaluation function children an empty set; explored← an empty set; frontier← Insert (Make-Node(Initial-state[problem],NULL,NULL,d,c),frontier) Loop do If Empty?(frontier) then return failure node ← POP(frontier) If Goal-Test[ problem] applied to State[node] succeeds thenreturn Solution(node) add State[node] to explored childrenExpand (node,problem) for each child in children If state [child] is not in explored or frontierthen frontierinsert(State [child], frontier) // sort frontier in ascending order of f-values else if child[state] is in frontier with higher f-values then replace that frontier node with child End Loop
Greedy best-first search • Let evaluation function f(n) be an estimate of cost from node n to goal • This function is often called a heuristic and is denoted by h(n). • f(n) = h(n) • e.g. hSLD(n) = straight-line distance from n to Bucharest • Greedy best-first search expands the node that appears to be closest to goal. • Contrast with uniform-cost search in which lowest cost path from start is expanded. • Heuristic function is the way knowledge about the problem is used to guide the search process.
Greedy best-first search Properties • Finds solution without ever expanding a node that is not on the solution path. • It is not optimal: the optimal path goes through Ptesti. • Minimizing h(n) is susceptible to false starts. • e.g. getting from Iasi to Fagaras: according to h(n), we take Neamt node to expand but it is a dead end. • If repeated states are not detected, the solution will never be found. Search gets stuck in loops: • Iasi →Neamet → Iasi → Neamet • The graph search version is Complete in finite spaces with repeated state checking but not in infinite ones.
A* search • Most widely known form of best-first search. • Key idea:avoid expanding paths that are already expensive. • Evaluation function:f(n) = g(n) + h(n) • g(n) = path cost so far to reach n. (used in Uniform Cost Search). • h(n) = estimated path cost to goal from n. (used in Greedy Best-First Search). • f(n) = estimated total cost of path through n to goal.
A* search • Definition: a heuristic h(n) is said to be admissible if it never overestimates the cost to reach the goal. h(n) h*(n) • where h*(n) is the TRUE cost from n to the goal. • e.g: hsld straight line can not be an overestimate. • Consequently: if h(n) is an admissible heuristic, then f(n) never overestimates the true cost of a solution through n. WHY? • It is true because g(n) gives the exact cost to reach n.
h*(n): true minimum cost to goal A* search root g(n):cost of path n h(n): Heuristic (expected) minimum cost to goal. (estimation) Goal
A* search • Theorem: When Tree-Search is used, A* is optimal if h(n) is an admissible heuristic. • Proof: • Let G be the optimal goal state reached by a path with cost : C* =f(G) = g(G). • Let G2 be some other goal state or the same state, but reached by a more costly path.
A* search f(G2) = g(G2)+h(G2) = g(G2) since h(G2) = 0 g(G2) > C* since G2 is suboptimal • Let n be any unexpanded node on the shortest path to the optimal goal G. f(n) = g(n) + h(n) ≤ C* since h is admissible Therefore, f(n) ≤ C* ≤ f(G2) • As a consequence, G2 will not be expanded and A* must return an optimal solution. • Example: the previous search: f(Bucharest)=450 was not chosen for expansion, even though Bucharest is the goal.
A* search • For Graph-Search, A* is optimal if h(n) is consistent. • Consistency (= Monotonicity): A heuristic is said to be consistent when for any node n, successor n’ of n, we have h(n) ≤ c(n,n’) + h(n’), where c(n,n’) is the (minimum) cost of a step from n to n’. • This is a form of triangular inequality: • Consistent heuristics are admissible. Not all admissible heuristics are consistent. • When a heuristic is consistent, the values of f(n) along any path are non decreasing. h(n) n c(n,n’) g h(n’) n’
A* search properties • Completeness: Yes, with f ≤ f(G). • Optimality: Yes. The tree-search version is optimal if h(n) is admissible, while the graph-search version is optimal if h(n) is consistent. • Time complexity: Exponential. • Space complexity: Keeps all nodes in memory.
Some admissible heuristics • 8-Puzzle: • g(n):the path cost can be measured by the total number of horizontal and vertical moves. • h(n): two different heuristics • h1 (n): number of misplaced tiles. • h2 (n): the sum of the distances of the tiles from their goal positions.
Local Search algorithms • Blind search and informed search strategies addressed a single category of problems: observable, deterministic, known environments where the solution is a sequence of actions. • The search algorithms we have seen so far keep track of the current state, the “frontier” of the search space, and the path to the final state. • In some problems, one doesn’t care about a solution path but only the final goal state. The solution is the goal state. • Example: 8-queen problem. • Local search algorithms are also useful for optimization problems where the goal is to find a state such that an objective function is optimized. • For the 8-queen algorithm, the objective function may be the number of attacks.
Local Search algorithms • Basic idea: • Local search algorithms operate on a single state – current state – and move to one of its neighboring states. • Therefore: Solution path does not need to be maintained. • Hence, the search is “local”. • Two advantages: • Use little memory. • More applicable in searching large/infinite search space. They find reasonable solutions in this case.
Local Search algorithms • A state space landscape is a graph of states associated with their costs • Problem: local search can get stuck on a local maximum and not find the optimal solution
Hill Climbing • Hill climbing search algorithm (also known as greedy local search) uses a loop that continually moves in the direction of increasing values (that is uphill). • Hill-climbing search modifies the current state to try to improve it, as shown by the arrow in figure on slide 28. • It terminates when it reaches a peak where no neighbor has a higher value. • A complete local search algorithm always find a goal if one exists. • An optimal algorithm always finds a global maximum/minimum.
Steepest ascent version Function Hill climbing (problem) return state that is a local maximum Inputs: problem, a problem Local variables: current, a node neighbor, a node Current ← Make-Node (initial-state [problem]) Loop do neighbor ← a highest-valued successor of current IfValue[neighbor] ≤ Value[current] then return state [current] Current ← neighbor
Simulated Annealing • Basic inspiration: What is annealing? • In metallurgy, annealing is the physical process used to temper or harden metals or glass by heating them to a high temperature and then gradually cooling them, thus allowing the material to coalesce into a low energy crystalline state. • Heating then slowly cooling a substance to obtain a strong crystalline structure. • Key idea: Simulated Annealing combines Hill Climbing with a random walk in some way that yields both efficiency and completeness. • Used to solve VLSI layout problems in the early 1980.
Local Beam Search • Unlike Hill Climbing, Local Beam Search keeps track of k states rather than just one. • It starts with k randomly generated states. • At each step, all the successors of all the states are generated. • If any one is a goal, the algorithm halts, otherwise it selects the k best successors from the complete list and repeats. • LBS≠ running k random restarts in parallel instead of sequence. • Drawback: less diversity → Stochastic Beam Search
Stochastic search: Genetic algorithms • Formally introduced in the US in the 70s by John Holland. • GAs emulate ideas from genetics and natural selection and can search potentially large spaces. • Before we can apply Genetic Algorithm to a problem, we need to answer: • How is an individual represented? • What is the fitness function? • How are individuals selected? • How do individuals reproduce?
Stochastic search: Genetic algorithms • Genetic algorithms is a variant of local beam search. • Successors in this case are generated by combining two parent states rather than modifying a single state. • Like local beam search, genetic algorithms starts with a set of k randomly generated states called Population. • Each state or individual is represented as a string over a finite alphabet. It is also called chromosome.
Stochastic search: Genetic algorithms • Each state is rated by the evaluation function called fitness function. • Fitness function should return higher values for better states. • For reproduction, individuals are selected with a probability which is directly proportional to the fitness score. • For each pair to be mated, a crossover point is randomly chosen from the positions in the string. • The offspring themselves are created by crossing over the parent strings at the crossover point. • Mutation is performed randomly with a small independent probability.
Summary • Informed search uses knowledge about the problem to reduce search costs. • This knowledge is expressed in terms of heuristics. • Best first search is a class of methods that use a variant of graph-search where the minimum-cost unexpanded nodes are chosen for expansion. • Best first search methods use a heuristic function h(n) that estimates the cost of a solution from a node. • Greedy best-first search is a best first search that expands nodes with minimal h(n). It is not optimal but often efficient. • A* search is a best first search that takes into account the total cost from the root node to goal node. It expands node with minimal f(n) = g(n) + h(n). It is complete and optimal provided that h(n) is admissible (for tree search) or consistent (for graph search). The space complexity is prohibitive.
Summary • Construction of heuristics can be done by relaxing the problem definition (in a sense simplifying the problem), by precomputing solution costs for subproblems or learning from experience with the problem class. • Local search methods keep small number of nodes in memory. They are suitable for problems where the solution is the goal state itself and not the path. • Hill climbing, simulated annealing and local beam search are examples of local search algorithms. • Stochastic algorithms represent another class of methods for informed search. Genetic algorithms are a kind of stochastic hill-climbing search in which a large population of states is maintained. New states are generated by mutation and by crossover which combines pairs of states from the population.