1 Data Structures and Algorithms Graphs I: Representation and Search Gal A. Kaminka Computer Science Department

2 Outline Reminder: Graphs Directed and undirected Matrix representation of graphs Directed and undirected Sparse matrices and sparse graphs Adjacency list representation

3 Graphs Tuple V is set of vertices E is a binary relation on V Each edge is a tuple, where v1,v2 in V |E| =< |V| 2

4 Directed and Undirected Graphs Directed graph: in E is ordered, i.e., a relation (v1,v2) Undirected graph: in E is un-ordered, i.e., a set { v1, v2 } Degree of a node X: Out-degree: number of edges In-degree: number of edges Degree: In-degree + Out-degree In undirected graph: number of edges { X, v2 }

5 Examples of graphs

6 Paths Path from vertex v 0 to vertex v k : A sequence of vertices For all 0 = exists. Path is of length k Two vertices x, y are adjacent if is an edge Path is simple if vertices in sequence are distinct. Cycle: if v 0 = v k is cycle of length 1

7 Connected graphs Undirected Connected graph: For any vertices x, y there exists a path x  y (= y  x) Directed connected graph: If underlying undirected graph is connected Strongly connected directed graph: If for any two vertices x, y there exist path x  y and path y  x Clique: a strongly connected component |V|-1 =< |E| =< |V| 2

8 Cycles and trees Graph with no cycles: acyclic Directed Acyclic Graph: DAG Undirected forest: Acyclic undirected graph Tree: undirected acyclic connected graph one connected component

9 Representing graphs Adjacency matrix: When graph is dense |E| close to |V| 2 Adjacency lists: When graph is sparse |E| << |V| 2

10 Adjacency Matrix Matrix of size |V| x |V| Each row (column) j correspond to a distinct vertex j “1” in cell if there is exists an edge Otherwise, “0” In an undirected graph, “1” in => “1” in “1” in means there’s a self-loop in vertex j

11 Examples

12 Adjacency matrix features Storage complexity: O(|V| 2 ) But can use bit-vector representation Undirected graph: symmetric along main diagonal A T transpose of A Undirected: A=A T In-degree of X: Sum along column X O(|V|) Out-degree of X: Sum along row X O(|V|) Very simple, good for small graphs Edge existence query: O(1)

13 But, …. Many graphs in practical problems are sparse Not many edges --- not all pairs x,y have edge x  y Matrix representation demands too much memory We want to reduce memory footprint Use sparse matrix techniques

14 Adjacency List An array Adj[ ] of size |V| Each cell holds a list for associated vertex Adj[u] is list of all vertices adjacent to u List does not have to be sorted Undirected graphs: Each edge is represented twice

15 Examples 1  3 2  2 3  1   2  3 2  1 3  1 4 

16 Adjacency list features Storage Complexity: O(|V| + |E|) In undirected graph: O(|V|+2*|E|) = O(|V|+|E|) Edge query check: O(|V|) in worst case Degree of node X: Out degree: Length of Adj[X] O(|V|) calculation In degree: Check all Adj[] lists O(|V|+|E|) Can be done in O(1) with some auxiliary information!

17 שאלות?

18 Graph Traversals (Search) We have covered some of these with binary trees Breadth-first search (BFS) Depth-first search (DFS) A traversal (search): An algorithm for systematically exploring a graph Visiting (all) vertices Until finding a goal vertex or until no more vertices Only for connected graphs

19 Breadth-first search One of the simplest algorithms Also one of the most important It forms the basis for MANY graph algorithms

20 BFS: Level-by-level traversal Given a starting vertex s Visit all vertices at increasing distance from s Visit all vertices at distance k from s Then visit all vertices at distance k+1 from s Then ….

21 BFS in a binary tree (reminder) BFS: visit all siblings before their descendents

22 BFS(tree t) 1. q  new queue 2. enqueue(q, t) 3. while (not empty(q)) 4. curr  dequeue(q) 5. visit curr // e.g., print curr.datum 6. enqueue(q, curr.left) 7. enqueue(q, curr.right) This version for binary trees only!

23 BFS for general graphs This version assumes vertices have two children left, right This is trivial to fix But still no good for general graphs It does not handle cycles Example.

24 Start with A. Put in the queue (marked red) ABGCEDF Queue: A

25 B and E are next ABGCEDF Queue: A B E

26 When we go to B, we put G and C in the queue When we go to E, we put D and F in the queue ABGCEDF Queue: A B E C G D F

27 When we go to B, we put G and C in the queue When we go to E, we put D and F in the queue ABGCEDF Queue: A B E C G D F

28 Suppose we now want to expand C. We put F in the queue again! ABGCEDF Queue: A B E C G D F F

29 Generalizing BFS Cycles: We need to save auxiliary information Each node needs to be marked Visited: No need to be put on queue Not visited:Put on queue when found What about assuming only two children vertices? Need to put all adjacent vertices in queue

30 BFS(graph g, vertex s) 1. unmark all vertices in G 2. q  new queue 3. mark s 4. enqueue(q, s) 5. while (not empty(q)) 6. curr  dequeue(q) 7. visit curr // e.g., print its data 8. for each edge 9. if V is unmarked 10. mark V 11. enqueue(q, V)

31 The general BFS algorithm Each vertex can be in one of three states: Unmarked and not on queue Marked and on queue Marked and off queue The algorithm moves vertices between these states

32 Handling vertices Unmarked and not on queue: Not reached yet Marked and on queue: Known, but adjacent vertices not visited yet (possibly) Marked and off queue: Known, all adjacent vertices on queue or done with

33 Start with A. Mark it. ABGCEDF Queue: A

34 Expand A’s adjacent vertices. Mark them and put them in queue. ABGCEDF Queue: A B E

35 Now take B off queue, and queue its neighbors. ABGCEDF Queue: A B E C G

36 Do same with E. ABGCEDF Queue: A B E C G D F

37 Visit C. Its neighbor F is already marked, so not queued. ABGCEDF Queue: A B E C G D F

38 Visit G. ABGCEDF Queue: A B E C G D F

39 Visit D. F, E marked so not queued. ABGCEDF Queue: A B E C G D F

40 Visit F. E, D, C marked, so not queued again. ABGCEDF Queue: A B E C G D F

41 Done. We have explored the graph in order: A B E C G D F. ABGCEDF Queue: A B E C G D F

42 Interesting features of BFS Complexity: O(|V| + |E|) All vertices put on queue exactly once For each vertex on queue, we expand its edges In other words, we traverse all edges once BFS finds shortest path from s to each vertex Shortest in terms of number of edges Why does this work?

43 Depth-first search Again, a simple and powerful algorithm Given a starting vertex s Pick an adjacent vertex, visit it. Then visit one of its adjacent vertices ….. Until impossible, then backtrack, visit another

44 DFS(graph g, vertex s) Assume all vertices initially unmarked 1. mark s 2. visit s // e.g., print its data 3. for each edge 4. if V is not marked 5. DFS(G, V)

45 Start with A. Mark it. ABGCEDF Current vertex: A

46 Expand A’s adjacent vertices. Pick one (B). Mark it and re-visit. ABGCEDF Current: B

47 Now expand B, and visit its neighbor, C. ABGCEDF Current: C

48 Visit F. Pick one of its neighbors, E. ABGCEDF Current: F

49 E’s adjacent vertices are A, D and F. A and F are marked, so pick D. ABGCEDF Current: E

50 Visit D. No new vertices available. Backtrack to E. Backtrack to F. Backtrack to C. Backtrack to B ABGCEDF Current: D

51 Visit G. No new vertices from here. Backtrack to B. Backtrack to A. E already marked so no new. ABGCEDF Current: G

52 Done. We have explored the graph in order: A B C F E D G ABGCEDF Current:

53 Interesting features of DFS Complexity: O(|V| + |E|) All vertices visited once, then marked For each vertex on queue, we examine all edges In other words, we traverse all edges once DFS does not necessarily find shortest path Why?