This assignment is due before 11:00PM on Wednesday, November 15, 2017. There are two parts to this homework assignment:

- A programming assignment on BFS / MSTs / Widest Path. This assignment also requires you to submit unit tests along with your implementation.
- A set of written questions about MSTs.

You’ll need to submit your solutions to both parts of the homework before the deadline.

Collaboration policy.

- You must do this assignment by yourself.
- You must never give or expose your solutions to an assignment to anyone who is taking the course. For example, you may not place your solutions in a public location (such as a website, a public code repository, or a printout left in a lab). If you leave your computer unattended, be sure to protect it with a password.
- You must never view someone else’s solutions to a programming assignment (or variant of an assignment). For example, you may not download solutions to a Coursera version of the assignment from the web. You may not use Coursera’s autograder to check the correctness of your solution.
- All solutions are checked with plagrism detection software. Any assignment that is flagged by the software will be automatically referred to the Office of Student Conduct, which will adjudicate whether the course collaboration policy was violated. The first violation will result in your overall course grade being decreased by one letter grade. A second violation will result in an F in the class.

We have put together several stub files, please ** click here to download**.

After de-compression, you should find a folder named “src” that contains seven Java files that you will need in this assignment.

- BFS.java
- Client.java
- Edge.java
- Graph.java
- Kruskal.java
- UnionFind.java
- WidestPath.java

You can think of the Internet as a graph where machines and routers are vertices, and the wires between them are edges. Various questions concern the implementation and structure of the Internet, such as “How many hops must a message take to reach a destination?” and “What is the cheapest way to connect and route all computers to each other?” In this assignment, you will implement several data structures and algorithms to answer these questions.

Note: using items from Java collections is okay (examples: `Collections.sort`

, `Queue`

interface), but we want you to write your own graph classes and graph algorithms, so please do not use the classes in `algs4.jar`

for this assignment.

Files to submit: Graph.java, GraphTest.java

Recall the graph lab, where we discussed graph representations. Here, we’ll implement an undirected, optionally-weighted graph. The graph should be immutable with respect to its vertices. Note that for an undirected graph, the edge from a vertex u to v is the same as the edge from vertex v to u.

We call the graph “optionally weighted” because it can be used by algorithms that use weights (like Kruskal’s) and by algorithms that do not (like BFS). An algorithm like BFS would simply ignore any weights present.

The implementation details are largely up to you, but be sure to consider the runtimes of each operation. In the Graph.java stub, we’ve provided target runtimes for each method you need to implement; be sure to keep these in mind! (Particularly, for anything that says expected or amortized, it is fair game to use a HashSet, HashMap or ArrayList on if you deem it to be necessary. These data structures offer expected-case O(1) find/insert/delete). Also, make sure that your solution only uses O(m + n) space - please don’t use adjacency matrices! Your solution does not need to handle self-loop or parallel edges (although you are welcome to), but it must support negative edge weights. Overall, the graph implementation is largely up to you for design as long as it meets the space and time requirements. This could include implementing an edge wrapper class if you chose to. The graph implementation also assumes that the vertices will be represented by integers that begin at 0.

Note that we’ve also provided a basic weighted edge class in `Edge.java`

with compareTo, equals, and hashCode implementations to make it easy to use for hashing and sorting.
This Edge class is tangential to the graph representation—it is just a wrapper for passing around edges, just as Graph.getEdges does (side note: the getEdges method should run in O(m) time). It will be useful in some later problems. It would probably be helpful for you to read through the Edge class to see how it works. DO NOT MODIFY `Edge.java`

!
Something else to keep in mind is what data your methods expose. Remember, you should practice defensive coding. If you return the data structures containing your graph to a client, then they can modify them as they wish (which is very bad!). Thus, you should use Collections.unmodifiableXXX to ensure that the user cannot modify the data structures you are passing them.

Files to submit: BFS.java, BFSTest.java

Recall the discussion of breadth-first search in lecture and in recitation. We want to implement BFS as a single-source shortest path algorithm, which we will use as a tool later in this homework assignment. The source and the target should be included in the resulting path. Remember even if the graph is weighted the BFS algorithm should not consider those edge weights. Also, your implementation should handle the unconnected case. The BFS.java comments contain necessary clarifying information, and you already know the algorithm, so give it your best shot!

Files to submit: UnionFind.java, UnionFindTest.java

Union-find is a data structure for tracking a set of elements partitioned into disjoint subsets. We can union these subsets together and then find which set an element belongs to. In particular, we can determine when two elements belong to the same set, which ends up being quite useful for Kruskal’s algorithm.

We want to start with the pointer-based implementation, in which each element keeps track of a “parent” pointer. At the beginning, everyone is their own parent, and two elements are in the same set if they share an ancestor. This allows for fast union operations, only involving changing a parent pointer. Think carefully, though, about what parent pointer that should be.

As with everything in CIS 121, there’s a tradeoff at play here. Fast union means slow find if we’re not careful, since find requires us to follow the parent pointers up to a root node. If our tree gets very tall, this could take a while. To alleviate this, we’ll implement the two optimizations discussed in lecture to help make find as fast as possible (and it turns out that that’s pretty fast!).

We recommend implementing this without any optimizations at first. Donald Knuth, responsible for popularizing big-Oh notation and inventing TEX, once famously said “Premature optimization is the root of all evil.” There are many ways to interpret this piece of wisdom, but here we take premature to mean “before it even works at all.” It’s also much easier to optimize working code when you have test cases to make sure you didn’t break anything.

The two optimizations are described in the following sections.

When we union two sets, one set essentially absorbs the other; the absorbing set continues to exist, but it’s the end of the road for the absorbed set. How, then, should we decide which set does the absorbing, and which set gets absorbed?

You’ll need to start keeping track of some property of your sets, and use that information to make a decision here. Remember that we’re trying to make find as fast as possible, so we want to keep our tree heights low; take this into consideration when deciding what property to track.

If you implement this optimization correctly, find() and union() will run in O(log n) time.

Another technique we can use to keep our traversals back to the root node from getting too long is by making them take just one hop. We call this path compression; whenever we traverse the tree to perform a find operation, we can just point the nodes along the traversal to the root. This should only take one or two lines of code to implement!

The amortized running time of find() and union() is very hard to reason about (take CIS 502 if you would like to grapple with that proof, or look here). Unbelievably, the path compression optimization makes the running time of those operations “effectively constant.” In reality, they are O(α(n)), where α(n) is the inverse Ackermann function. α(n) grows so slowly that for almost any practical value of n, α(n) < 5. Hence, “effectively constant”! That’s pretty cool.

Remember to implement both optimizations in the same method. Also, when implementing path compression keep in mind that rank is not necessarily equal to depth. Finally for consistency, during the union step for u and v, if i is the representative of u and j is the representative for v make i the parent of j if both i and j have the same rank.

Files to submit: Kruskal.java, KruskalTest.java

Kruskal’s algorithm computes the minimum spanning tree of a graph by using the union-find data structure. You are to implement this algorithm, which is further described by the comments in Kruskal.java. For this algorithm you are allowed to use Collections.Sort to sort the edge weights. Furthermore, if the edge weights are not distinct you can return any valid MST.

Files to submit: WidestPath.java, WidestPathTest.java

Consider a graph and a path P. The path P is a widest path when the weight of smallest edge in P is maximal. This maximal minimum-weight edge is known as a bottleneck edge. In other words, if you wanted to transfer a large amount of something (internet packets, car traffic, etc.) along a graph, and you considered edge weights to be capacities, you would want to find a path with the largest overall capacity. If we think about Internet packets, then the overall capacity of a route would be limited by the lowest-bandwidth connection in that route, as we would not be able to send more information along that path. Thus, the widest path is the path along which we can send the most stuff. The constraining edge on that path, or in this example, the lowest-bandwidth connection, would be the bottleneck edge of the path.

How can we algorithmically find a graph’s widest path? That’s up to you to devise and implement. There are multiple ways to solve this problem, so consider how you can do it with what you’ve already built (namely, BFS and Kruskal’s). Please do this without modifying the behavior of the BFS and Kruskal’s implementations you have already built. Note that within a widest path, edge weights are not necessarily distinct, and those weights take precedence over the length of the path. To clarify, if there are multiple widest paths with the same bottleneck but different lengths then any of those paths are a valid solution. Runtime for this algorithm should be in O(mlgn).

Files to submit: Client.java, ClientTest.java.

Let us take a step back. We have spent a lot of time in this class implementing APIs that do cool things. In this homework, we will briefly use an API we have developed as a code client. A brief aside: API stands for Application Program Interface. In the context of the web, it often refers to a set of URLs and requests that one can use to interact with a third-party service; for example, Candy Crush integrates with the Facebook API. More generally, it refers to the points of contact between any two independent pieces of a software system. In our context, this means the public methods in a Java class or interface that you are allowed to call. For example, the WidestPath.java API is simply WidestPath.getWidestPath().

Now, imagine that you work for an Internet service provider (ISP). Imagine also that you have a list of all direct links between routers on your company’s network, and you know the bandwidth capacity of each link. As an ISP, you’re concerned about effectively routing Internet traffic. Note: IP addresses are in the standard format of IP addresses (i.e. XXX.XXX.XXX.XXX, where the number of digits in each section is from 1 to 3) which can be stored as strings.

Next, consider strategies one might use in choosing a path between two routers. Some common metrics include geographic distance (weighted shortest path) and number of hops (unweighted shortest path). Your company, however, wants to consider widest paths between routers in order to maximize available bandwidth. You have the WidestPath.java API available, so you do not need to implement a solution; instead, you need to marshall your own data into the format of the API and convert the result of the API call into a usable answer. Figuring out how to do this is the challenge of part six.

The above parts together are worth a total of 85 points. The remaining 15 points are awarded for code style, documentation, and sensible tests. Style is worth 5 of the 15 points, and you will be graded according to the 121 style guide.

You will need to write comprehensive test cases for each of the methods you implement. This assignment in particular is highly compositional, with later parts building heavily on earlier parts, so we’d like to stress the importance of doing bottom-up testing for each component before using it elsewhere. We also encourage you to work out small examples of BFS, Kruskal’s, and Widest Path by hand in order to write unit tests.

Make sure you consider edge cases, i.e., the more “interesting” inputs and situations. Use multiple methods instead of cramming a bunch of asserts into a single test method. Be sure to demonstrate that you considered “bad” inputs such as null inputs. Be sure to also demonstrate that you’ve tested for inputs to methods that should throw exceptions! Your test cases will be manually graded by the TAs, and are worth 10 of the 15 points. You will have to thoroughly test your code to get full points! This includes testing any helper methods you have written. Note: you will not be able to write JUnit test cases for any private methods. Instead, make them package-private by leaving off any privacy modifier (i.e., do not write public or private).

The goals of this assignment are to test your understanding of the material covered in sections 1.5 and 4.1-4.3 of the textbook, and the lecture and recitation materials. You should read the textbook chapters before doing this part of the assignment.

Written homeworks must be typeset in LaTeX and submitted in PDF format. Please insert a page break between each question, so that your answer to each question starts on a new page in your PDF document.

Please answer the following questions (all running times should have a theta bound):

- What is the running time of your implementation of Kruskal’s algorithm? Was there any part of the code that acted as a bottleneck? If so, how could that be avoided?
- What is the running time of your implementation of Widest Path? Can you think of any alternative ways to solve the Widest Path problem?
- What were some of the advantages and disadvantages of the Graph class using ints to represent vertices? Consider this question in the context of Client.java.
- Where possible, we used a functional structure in this homework, encouraging you to avoid statefulness. For example, the BFS.getShortestPath() method is static and takes in the graph as an argument, `a la OCaml. Discuss some pros and cons of this kind of structuring.

True or False: Given any two connected components that are generated as Kruskal’s algorithm is running (but before it has completed), the smallest edge connecting those two components is part of the MST. Explain your answer.

Prove the following, known as the cycle property: Given any cycle in an edge-weighted graph (all edge weights distinct), the edge of maximum weight in the cycle does not belong to the MST of the graph.

A bridge in a graph is an edge that, if removed, would increase the number of connected components. A graph that has no bridges is said to be two-edge connected. Develop a linear-time DFS-based algorithm to determine all bridges in a graph and whether it is 2-edge connected.

An MST edge whose deletion from the graph would cause the MST weight to increase is called a critical edge. Show how to find all critical edges in a graph in time proportional to E log E. Note: This question assumes that edge weights are not necessarily distinct (otherwise all edges in the MST are critical). *Hint:* Apply Q4 and show that an edge e is critical if and only if it is a bridge in the subgraph containing all edges with weight less than or equal to the weight of edge e.