CS 481 Midterm
Backtracing
"Goal Directed" reasoning backwards to a current state from a goal state. This is done by creating sub goal states on the way to current goal state.
Heuristic advantages
- Can find solutions that are complex and time-consuming for exact methods - Can often find good solutions, that are probably not optimal but better than uninformed methods
Heuristic disadvantages
- Heuristics may not find optimal solution - Heuristics are hard to make, and a bad heuristic leads to suboptimal solutions or loops
DFS disadvantages
- May be less efficient - May take a longer path to goal state - Goes deep down searching and sometime it may go to the infinite loop
Finding a goal state
A control strategy determines how transformations between states take place.
Iterative Deepening Search
A depth limited search that gradually increases the limit. Generates states multiple times. IE DFS with depth limit that increases overtime: Aims to take advantages of DFS with search type and BFS with tree limit.
Goal Space
A state where the agent may end the search.
Fully observable
An agent's sensors give it access to the complete state of the environment at each point in time.
BFS time complexity
BFS may be quicker, but need to make space for every node connected top root n1, and every node at n2 if not found.
GBFS advantages
Can find path very quickly if heuristic is good.
disadvantages of hill climbing
Can get stuck at a suboptimal end state, may terminate immediately
A* advantages
Combines UCS and GBFS together to get a better overall cost.
Minimax Disadvantages
Computentalizally expensive, difficult to implement
Discrete
Consists of a finite number of states and agents with a finite number of actions. Example: Tic-Tac-Toe Board. Move amount may vary but there is an end.
A* Disadvantages
Costly in both time and space
DFS Space complexity
DFS has a smaller space cost than bfs.
Episodic
Each game state is independent of each other. The states have no bearing on each other. Example: A machine that processes images, each image has no influence over another.
Dynamic
Environment could be changing while the agent is preempting or performing action. Example stock market, or Peggle with the extra ball platform.
Static
Environment is completely unchanged while agent is preempting a move. Example chess without a clock or crossword.
Sequential
Environment where next state is dependent on current action. Example poker or chess.
Breath First Search (BFS)
Expand and evaluate the shallowest node first. Traversing level by level. Goal is to find shortest path depth wise but not least costly
Depth First Search (DFS)
Expand the deepest node first. Backtrace.
Minimax complexity
Exponential deptj of game tree
BFS data structure
FIFO Queue
UCS advantages
Finds lowest cost solution
Uniform Cost Search (UCS)
Finds the cheapest path to a goal state. With a visited list. Traverses by calculating the total costs of paths before traversing (no evaluation of states where cost is being considered until it is the cheapest), always choosing the cheapest path.
BFS space complexity
Has to make space for all connecting nodes, space cost high
Hill climbing
Hill climbing is a tech for finding local max or min of a function. Used for optimization and machine learning.
Greedy Best First Search
Informed Search Algorithm, Greed always takes the least cost with a partially informed graph. Example: Having a train map with the straight line distance to goal state calculated in node. Has elements of UCS but instead of keeping track of total cost, it keeps track of total cost to goal.
DFS data structure
LIFO Stack
Minimax Search Algorithm
Max, min, fight. Minimum of the maximum. Alpha beta pruning is disregarding any value that is lower than maximum. V= total for given branch that is accepteddzz, Alpha best total for tree, pass beta down tree checking if there is anything higher on the tree to disreguard.
Greedy Best First Search disadvantages.
Not promised to finish quickly, often does not find the shortest path. g(n)
DFS time complexity
O(V+E) V=num of vertices, E=num of edges. Visits all vertices recursively.
Stochastic
Outcomes are not certain Randomness changes the system dynamics. Outcomes are uncertain.
continuous environment
Percepts, actions and episodes are continuous (think robot car). Environment is inconsistent and does not have a finite amount of agents, actions, or states
Partially Observable
Players cannot see the entire game state, such as in a game of poker
GBFS
Priority queue
UCS Structure
Priority queue
PL to CNF
Remove bicondition, example: A<-->B to (A->B) ^ (B->A) Remove implication: A->B to ~AVB Move negation inwards: ~(AVB) to ~A^~B ~(AVB) to ~A^~B ~(~A) to A A^(BVC) to (A^B) V(A^C)
Advantages of hill climbing
Simple to implement, efficient for larger complex problems
UCS disadvantages
Takes alot of time and space
Graph search
Tech for finding a path from one node in a graph to another. Used primarily for routing and planning
deterministic
The outcome of the game is determined purely by the current game state and players' actions. Produce the exact results for a set of inputs.
Known
The outcomes (or probabilities) for all actions are given
unknown
The outcomes are uncertain for the environment and not given. Agent gathers info about given environment and is not given info forthright.
Comp
The players compete against each other to achieve own goals
Coop
The players work together to achieve a common goal
State space
The representation of all possible game states the game can be in.
Multiple agents
There are multiple players in the game
Single agent
There is only one player in the game
forward reasoning
Using data and reasoning driven approach to start from current state and find goal state.
A* Star Algorithm
Using the cost of an informed heuristic like greedy search, but adds the heuristic with cost of total edges to n. g(n+n2+...) + h(n)
Minimax Advantages
complete algorithm, will find best solution given infinite amount of time
UCS structure
priority queue with visited