550 likes | 636 Views
Uninformed Search. Problem-solving agents. Example: Romania. On holiday in Romania; currently in Arad . Flight leaves tomorrow from Bucharest. What do we need to define?. Problem Formulation. The process of defining actions, states and goal. States:
E N D
Example: Romania • On holiday in Romania; currently in Arad. • Flight leaves tomorrow from Bucharest
Problem Formulation • The process of defining actions, states and goal. • States: • Cities (e.g. Arad, Sibiu, Bucharest, etc) • Actions: • GoTo(adjacent city) • Goal: • Bucharest Why not “turn left 5 degrees” or “walk 100 meters forward…”?
Abstraction • The process of removing details from a representation. • Simplifies the problem • Makes problems tractable (possible to solve) • Humans are great at this! • Imagine a hierarchy in which another agent takes care of the lower level details, such as navigating from the city center to the highway.
Back to Arad… • We are in Arad and need to find our way to Bucharest.
Step 1 – Check Goal Condition • Check, are we at the goal? • (obviously not in this case, but we need to check in case we were)
Step 2 – Expand Current Node • Enumerate all the possible actions you could take from the current state • Formally: apply each legal action to the current state, thereby generating a new set of states. • From Arad can go to: • Sibiu • Timisoara • Zerind
Step 3 – Select which action to perform • Perform one of the possible actions(e.g. GoTo(Sibiu)) • Then go back to Step 1 and repeat.
This is an example of Tree Search • Exploration of state space by generating successors of already-explored states (a.k.a. expanding states) • Usually performed offline, as a simulation • Returns the sequence of actions that should be performed to reach the goal, or that the goal is unreachable.
Tree search example: need to process the descendants of Sibiu Note that we can loop back to Arad. Have to make sure we don’t go in circles forever!
Implementation: general tree search a.k.a. frontier
This is the part that distinguishes different search strategies
Search strategies • A search strategy is defined by picking the order of node expansion
Uninformed search strategies • Uninformed search strategies use only the information available in the problem definition • What does it mean to be uninformed? • You only know the topology of which states are connected by which actions. No additional information. • Later we’ll talk about informed search, in which you can estimate which actions are likely to be better than others.
Breadth-first search • Expand shallowest unexpanded node • Implementation: • Fringe is a FIFO queue, i.e., new successors go at end
Breadth-first search • Expand shallowest unexpanded node • Implementation: • Fringeis a FIFO queue, i.e., new successors go at end
Breadth-first search • Expand shallowest unexpanded node • Implementation: • Fringeis a FIFO queue, i.e., new successors go at end
Breadth-first search • Expand shallowest unexpanded node • Implementation: • Fringeis a FIFO queue, i.e., new successors go at end
Search Strategy Evaluation: finding solutions • Strategies are evaluated along the following dimensions: • completeness: does it always find a solution if one exists? • optimality: does it always find a least-cost solution?
Search Strategy Evaluation: complexity (cost) • Two types of complexity • time complexity: number of nodes visited • space complexity: maximum number of nodes in memory • Time and space complexity are measured in terms of • b: maximum branching factor of the search tree (may ∞) • d: depth of the least-cost solution • m: maximum depth of the state space (may be ∞)
Properties of breadth-first search • Complete? • Yes (if b is finite) • Optimal? • Yes (if cost = 1 per step) • Time? • 1+b+b2+b3+… +bd= O(bd) • Space? • O(bd) (keeps every node in memory)
Problems of breadth first search • Space is the biggest problem (more than time) • Example from book, BFS b=10 to depth of 10 • 3 hours (not so bad) • 10 terabytes of memory (really bad) • Only reason speed is not a problem is you run out of memory first
Problems of breadth first search • BFS is not optimal if the cost of some actions is greater than others…
Uniform-cost search • For graphs with actions of different cost • Equivalent to breadth-first if step costs all equal • Expand least-cost unexpanded node • Implementation: • fringe = queue sorted by path cost g(n), from smallest to largest (i.e. a priority queue)
Uniform-cost search • Complete? • Yes, if step cost ≥ ε • Time? • 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) See book for detailed analysis.
Depth-first search • Expand deepest unexpanded node • Implementation: • fringe = LIFO queue, i.e., put successors at front (i.e. a stack)
Depth-first search • Expand deepest unexpanded node • Implementation: • fringe = LIFO queue, i.e., put successors at front
Depth-first search • Expand deepest unexpanded node • Implementation: • fringe = LIFO queue, i.e., put successors at front
Depth-first search • Expand deepest unexpanded node • Implementation: • fringe = LIFO queue, i.e., put successors at front
Depth-first search • Expand deepest unexpanded node • Implementation: • fringe = LIFO queue, i.e., put successors at front
Depth-first search • Expand deepest unexpanded node • Implementation: • fringe = LIFO queue, i.e., put successors at front
This is the part that distinguishes different search algorithms
Search Solution • Each node needs to keep track of its parent • Once the goal is found, traverse up the tree to the root to find the solution
Properties of depth-first search • Complete? • No: fails in infinite-depth spaces • Yes: in finite spaces • Optimal? • No • Time? • O(bm): (m is max depth of state space) • terrible if m is much larger than d • but if solutions are plentiful, may be much faster than breadth-first • Space? • O(bm), i.e., linear space!
Depth-limited search • depth-first search with depth limit l (i.e., don’t expand nodes past depth l) • … will fail if the goal is below the depth limit