Personal Solutions to David Barber’s Bayesian Reasoning and Machine Learning (Chapter 2)

Exercise 2.1.

Consider an adjacency matrix A with elements \([A]_{i,j} = 1\) if one can reach state \(i\) from state \(j\) in one timestep, and \(0\) otherwise.

Show that the matrix \([A]_{i,j}^k\) represents the number of paths that lead from state \(j\) to \(i\) in \(k\) timesteps. Hence derive an algorithm that will find the minimum number of steps to get from state \(j\) to state \(i\).

Ans:

An example that satisfies (but does not prove) \([A]_{i,j}^k\) is \(fig\ 2.2(a)\).

In order to find the minimum number of steps to get from state \(j\) to state \(i\), we should increase \(k\) in calculating \([A]^k\), such that \([A]^k_{i,j}\) becomes non-zero. \(k\) in that case will be the smallest number of steps needed to get from state \(j\) to state \(i\).

Exercise 2.2.

For an \(N \times N\) symmetric adjacency matrix \(A\), describe an algorithm to find the connected components. You may wish to examine connectedComponents.m.

Ans:

The algorithm is as follows:

Start with an empty collection of connected components. For each row of the adjacency matrix, find the nodes that are connected. Iterate through the collection of connected compontents and see if there’s a match. If there is none, put the row’s connected components as a new entry. If there are matches, take the union of the matches along with the connected components specified by the current row. Remove the components that have a match, and then add the unioned components (as one big component) into the connected components collection. At the end, you should have a collection of components, where each component has nodes that are linked to each other.

See my implementation here and tests here

Exercise 2.3.

Show that for a connected graph that is singly-connected, the number of edges \(E\) must be equal to the number of nodes minus \(1\), \(E = V - 1\). Give an example graph with \(E = V - 1\) that is not singly-connected. Hence the condition \(E = V - 1\) is a necessary but not sufficient condition for a graph to be singly-connected.

Ans

Need to prove that All singly-connected graphs have the property \(E = V - 1\).

# edges # vertices E = V - 1 example
0 1 true x
1 2 true x—-y
2 3 true x—-y—-z

Base case:

Let \(v\) be the number of vertices. Let \(v = 1\). In this case, the number of edges must be \(0\). \(0 = 1 - 1\), so the base case satisfies \(E = V - 1 \).

Assumption step:

Let \(V\) be any arbitrary number \(k\). Let’s assume that when \(V\) is \(k\), the number of edges is \(k-1\). In other words:

Inductive step:

Let \(V = k + 1 \):

Thus all singly-connected graphs must have the property \(E = V - 1\).

An example of a non-singly-connected graph that satisfies \(E = V - 1\) is the following:

Non-singly connected graph
Non-singly connected graph

In this particular case, we have two components. One component is cyclical (thus not singly-connected) while the other is an island. Thus \(E = V - 1\) property is necessary but not sufficient to prove that a graph is singly-connected.

Example 2.4

Describe a procedure to determine if a graph is singly-connected.

Ans

A graph is singly-connected if any node can reach another (or itself) only one way. An algorithm to determine a graph’s singly-connectedness is by using some tree-traversing algorithm (such as Depth-First Search or Breadth-First Search).

  1. Keep a tally of visited nodes.
  2. Traverse the tree, starting at a node.
  3. At each node, see if all the adjacent nodes have been visited before.
    • If it has, then there is a cycle, so it must not be a singly-connected graph. Return.
    • If it hasn’t, just add the approached node to the list of visited nodes.
    • Keep on going until all nodes have been traversed, or until a cycle has been discovered.

Example 2.5

Describe a procedure to determine all the ancestors of a set of nodes in a DAG.

Ans

For each node in the set of nodes, keep a list of “branches” that lets a root node reach each node in that set of nodes. Starting from a node in the set of nodes, we traverse the parent recursively until we reach a root node. When we reach a root node, then we have one full traversed branch. We keep a running list of traversed branches. We are done once we’ve traversed all the paths up to root node(s) for each node in the nodes of interest.

Example 2.6

WikiAdjSmall.mat contains a random selection of \(1000\) Wiki authors, with a link between two authors if they ‘know’ each other (see snap.stanford.edu/data/wiki-Vote.html). Assume that if \(i\) ‘knows’ \(j\), then \(j\) ‘knows’ \(i\). Plot a histogram of the separation (the length of the path between two users on the graph corresponding to the adjacency matrix) between all users based on separations from \(1\) to \(20\). That is the bin \(n(s)\) in the histogram contains the number of pairs with separation \(s\).

Ans

The problem is essentially asking us to compute \(A^k\) where \(k\) is in the range \([1..20]\).

lets us know how many paths one could take in order to get from \(i\) to \(j\) in \(k\) steps.

When calculating the histogram, if \(A^k_{i,j}\) is greater than zero, then we increase the \(k^{th}\) bin. Otherwise, we don’t increase the latter.

Degrees of Separation Among Wikipedia Authors
Degrees of Separation Among Wikipedia Authors

The S-shaped curve does not look surprising. The graph is most likely cyclical and thus an arbitrarily-chosen author \(i\) and arbitrarily-chosen author \(j\) might have several degrees of separation. For example, let’s say that \(i\) and \(j\) are separated by one edge. It’s also very likely that they might know someone in common, author \(l\), so that \(i\) and \(j\) can also be reached in two steps, etc.

Example 2.7

The file cliques.mat contains a list of \(100\) cliques defined on a graph of \(10\) nodes. Your task is to return a set of unique maximal cliques, eliminating cliques that are wholly contained within another. Once you have found a clique, you can represent it in binary form as, for example (\(1110011110\)) which says that this clique contains variables \(1, 2, 3, 6, 7, 8, 9\), reading from left to right. Converting this binary representation to decimal (with the rightmost bit being the units and the leftmost \(29\)) this corresponds to the number \(926\). Using this decimal representation, write the list of unique cliques, ordered from lowest decimal representation to highest. Describe fully the stages of the algorithm you use to find these unique cliques. Hint: you may find examining uniquepots.m useful.

Ans

I was able to find 15 unique maximal cliques. They are

When sorted in decimals, they are: \(119\), \(447\), \(463\), \(487\), \(703\), \(751\), \(755\), \(765\), \(831\), \(863\), \(886\), \(893\), \(954\), \(983\), \(1006\).

My steps for finding the unique maximal cliques were as follows.

 For each clique,
   see if is a subset of another.
   if not a subset, then
     it must be a maximal clique.
     store this clique.

 To see if a clique is a subset of another,
    Take the intersections of the two cliques.
    If the intersection length is the same as
      the current clique, and
      the clique we are comparing to is not
      the same as the current clique, then
      the current clique is a subset
      (and is not a maximal clique.)

For more details, see this notebook

Example 2.8

Explain how to construct a graph with \(N\) nodes, where \(N\) is even, that contains at least \((N/2)^2\) maximal cliques.

Ans

When there are two nodes, it is very simple. Connecting the two nodes satisfies the \((N/2)^2\) maximal cliques constraint, where the maximal clique length for all cliques is 1.

When there are more than two nodes, we create an \(N\)-gon. In the case of \(4\) nodes, we create a square, and that is enough to satisfy the constraint.

\(N\) \((N/2)^2\)
2 1
4 4
6 9
8 16

Having even number of nodes greater than four makes the arrangement more interesting. We still create \(N\)-gons, but we also connect nodes to form trapezoids inside the \(N\)-gons. See the third and fourth example below.

Examples of graphs that satisfy the maximal cliques constraint
Examples of graphs that satisfy the maximal cliques constraint

Exercise 2.9

Let \(N\) be divisible by \(3\). Construct a graph with \(N\) nodes by partitioning the nodes into \(N/3\) subsets, each subset containing \(3\) nodes. Then connect all nodes, provided they are not in the same subset. Show that such a graph has \(3N/3\) maximal cliques. This shows that a graph can have an exponentially large number of maximal cliques.

Ans

When \(N\) is greater than \(3\) and is divisible by \(3\), for any arbitrary graph, it seems that we will have \(3N/3\) maximal cliques. One can see that the second and third examples below show the exponential growth.

Exponential growth of maximal cliques
Exponential growth of maximal cliques

Exercise 2.10

A jewel was stolen from a room during a party. Each of the guests (\(A\),\(B\),\(C\),\(D\),\(E\),\(F\)) enters the room, stays for some time, and leaves. The testimonies they gave to the police are:

  1. \(A\): I was present in the room with \(E\), \(B\)
  2. \(B\): I was present in the room with \(A\), \(F\) and \(E\)
  3. \(C\): I was present in the room with \(F\) and \(D\)
  4. \(D\): I was present in the room with \(A\) and \(F\)
  5. \(E\): I was present in the room with \(C\)
  6. \(F\): I was present in the room with \(C\) and \(E\)

One of the statements in the testimonies is false. Which is it?

Notes:

If any two people are present in the room at the same time, then at least one of them will report the fact that he was present in the room with the other person. Each above testimony should be interpreted as, for example:

\(A\): I was present in the room with \(E\). I was present in the room with \(B\). It does not necessarily mean that A was simultaneously present in the room with both \(E\) and \(B\) (so \(A\) and \(B\) could have been present in the room at some time \(t_1\), and \(A\) and \(E\) present in the room at some other time \(t_2\). The other testimonies have a similar interpretation.

The testimonies could be partially false. If \(X\) says that he was present in the room with \(Y\) and \(Z\), it could be that the statement \(X\) was present in the room with \(Y\) is true, but X was present in the room with \(Z\) is false.

All testimonies
All testimonies

The direction of an arrow depicts who is claiming to have been with whom. For example, \(A\) pointing to \(B\) means that \(A\) claimed to be with person \(B\).

Since we know that only one statement is false, we can assume that bidirectional edges are true (e.g. both \(A\) and \(B\) were definitely together). Both statements have to be true because if one person were lying, then the other person must have lied too, which violates the constraint that only one statement is false. Thus, we know that \(A\) must have been in the room with \(B\), and \(F\) and \(C\) must also have been together.

D lied about being with A and F
D lied about being with A and F

We know that \(D\) lied about being with \(A\) and \(F\) because if \(D\) was actually with \(A\) and \(F\), then \(D\) would have to have been in the room with \(B\) and \(E\) as well, but \(D\) did not say anything about being with those two, and vice versa. Therefore, we should be more suspicious of \(D\).