# Visualizing an Assassin Puzzle

Over at Math3ma, Tai-Danae Bradley shared the following puzzle, which she also featured in a fantastic (spoiler-free) YouTube video. If you’re seeing this for the first time, watch the video first.

Consider a square in the xy-plane, and let A (an “assassin”) and T (a “target”) be two arbitrary-but-fixed points within the square. Suppose that the square behaves like a billiard table, so that any ray (a.k.a “shot”) from the assassin will bounce off the sides of the square, with the angle of incidence equaling the angle of reflection.

Puzzle: Is it possible to block any possible shot from A to T by placing a finite number of points in the square?

This puzzle found its way to me through Tai-Danae’s video, via category theorist Emily Riehl, via a talk by the recently deceased Fields Medalist Maryam Mirzakhani, who studied the problem in more generality. I’m not familiar with her work, but knowing mathematicians it’s probably set in an arbitrary complex $n$-manifold.

See Tai-Danae’s post for a proof, which left such an impression on me I had to dig deeper. In this post I’ll discuss a visualization I made—now posted at the end of Tai-Danae’s article—as well as here and below (to avoid spoilers). In the visualization, mouse movement chooses the firing direction for the assassin, and the target is in green. Dragging the target with the mouse updates the position of the guards. The source code is on Github.

## Outline

The visualization uses d3 library, which was made for visualizations that dynamically update with data. I use it because it can draw SVGs real nice.

The meat of the visualization is in two geometric functions.

1. Decompose a ray into a series of line segments—its path as it bounces off the walls—stopping if it intersects any of the points in the plane.
2. Compute the optimal position of the guards, given the boundary square and the positions of the assassin and target.

Both of these functions, along with all the geometry that supports them, is in geometry.js. The rest of the demo is defined in main.js, in which I oafishly trample over d3 best practices to arrive miraculously at a working product. Critiques welcome 🙂

As with most programming and software problems, the key to implementing these functions while maintaining your sanity is breaking it down into manageable pieces. Incrementalism is your friend.

## Vectors, rays, rectangles, and ray splitting

We start at the bottom with a Vector class with helpful methods for adding, scaling, and computing norms and inner products.

function innerProduct(a, b) {
return a.x * b.x + a.y * b.y;
}

class Vector {
constructor(x, y) {
this.x = x;
this.y = y;
}

normalized() { ... }
norm() { ... }
subtract(vector) { ... }
scale(length) { ... }
distance(vector) { ... }
midpoint(b) { ... }
}


This allows one to compute the distance between two points, e.g., with vector.subtract(otherVector).norm().

Next we define a class for a ray, which is represented by its center (a vector) and a direction (a vector).

class Ray {
constructor(center, direction, length=100000) {
this.center = center;
this.length = length;

if (direction.x == 0 && direction.y == 0) {
throw "Can't have zero direction";
}
this.direction = direction.normalized();
}

endpoint() {
}

intersects(point) {
let shiftedPoint = point.subtract(this.center);
let signedLength = innerProduct(shiftedPoint, this.direction);
let projectedVector = this.direction.scale(signedLength);
let differenceVector = shiftedPoint.subtract(projectedVector);

if (signedLength > 0
&& this.length > signedLength
} else {
return null;
}
}
}


The ray must be finite for us to draw it, but the length we've chosen is so large that, as you can see in the visualization, it's effectively infinite. Feel free to scale it up even longer.

The interesting bit is the intersection function. We want to compute whether a ray intersects a point. To do this, we use the inner product as a decision rule to compute the distance of a point from a line. If that distance is very small, we say they intersect.

In our demo points are not infinitesimal, but rather have a small radius described by intersectionRadius. For the sake of being able to see anything we set this to 3 pixels. If it’s too small the demo will look bad. The ray won’t stop when it should appear to stop, and it can appear to hit the target when it doesn’t.

Next up we have a class for a Rectangle, which is where the magic happens. The boilerplate and helper methods:

class Rectangle {
constructor(bottomLeft, topRight) {
this.bottomLeft = bottomLeft;
this.topRight = topRight;
}

topLeft() { ... }
center() { ... }
width() { .. }
height() { ... }
contains(vector) { ... }


The function rayToPoints that splits a ray into line segments from bouncing depends on three helper functions:

1. rayIntersection: Compute the intersection point of a ray with the rectangle.
2. isOnVerticalWall: Determine if a point is on a vertical or horizontal wall of the rectangle, raising an error if neither.
3. splitRay: Split a ray into a line segment and a shorter ray that’s “bounced” off the wall of the rectangle.

(2) is trivial, computing some x- and y-coordinate distances up to some error tolerance. (1) involves parameterizing the ray and checking one of four inequalities. If the bottom left of the rectangle is $(x_1, y_1)$ and the top right is $(x_2, y_2)$ and the ray is written as $\{ (c_1 + t v_1, c_2 + t v_2) \mid t > 0 \}$, then—with some elbow grease—the following four equations provide all possibilities, with some special cases for vertical or horizontal rays:

\displaystyle \begin{aligned} c_2 + t v_2 &= y_2 & \textup{ and } \hspace{2mm} & x_1 \leq c_1 + t v_1 \leq x_2 & \textup{ (intersects top)} \\ c_2 + t v_2 &= y_1 & \textup{ and } \hspace{2mm} & x_1 \leq c_1 + t v_1 \leq x_2 & \textup{ (intersects bottom)} \\ c_1 + t v_1 &= x_1 & \textup{ and } \hspace{2mm} & y_1 \leq c_2 + t v_2 \leq y_2 & \textup{ (intersects left)} \\ c_1 + t v_1 &= x_2 & \textup{ and } \hspace{2mm} & y_1 \leq c_2 + t v_2 \leq y_2 & \textup{ (intersects right)} \\ \end{aligned}

In code:

  rayIntersection(ray) {
let c1 = ray.center.x;
let c2 = ray.center.y;
let v1 = ray.direction.x;
let v2 = ray.direction.y;
let x1 = this.bottomLeft.x;
let y1 = this.bottomLeft.y;
let x2 = this.topRight.x;
let y2 = this.topRight.y;

// ray is vertically up or down
if (epsilon > Math.abs(v1)) {
return new Vector(c1, (v2 > 0 ? y2 : y1));
}

// ray is horizontally left or right
if (epsilon > Math.abs(v2)) {
return new Vector((v1 > 0 ? x2 : x1), c2);
}

let tTop = (y2 - c2) / v2;
let tBottom = (y1 - c2) / v2;
let tLeft = (x1 - c1) / v1;
let tRight = (x2 - c1) / v1;

// Exactly one t value should be both positive and result in a point
// within the rectangle

let tValues = [tTop, tBottom, tLeft, tRight];
for (let i = 0; i  epsilon && this.contains(intersection)) {
return intersection;
}
}

throw "Unexpected error: ray never intersects rectangle!";
}


Next, splitRay splits a ray into a single line segment and the “remaining” ray, by computing the ray’s intersection with the rectangle, and having the “remaining” ray mirror the direction of approach with a new center that lies on the wall of the rectangle. The new ray length is appropriately shorter. If we run out of ray length, we simply return a segment with a null ray.

  splitRay(ray) {
let segment = [ray.center, this.rayIntersection(ray)];
let segmentLength = segment[0].subtract(segment[1]).norm();
let remainingLength = ray.length - segmentLength;

if (remainingLength < 10) {
return {
segment: [ray.center, ray.endpoint()],
ray: null
};
}

let vertical = this.isOnVerticalWall(segment[1]);
let newRayDirection = null;

if (vertical) {
newRayDirection = new Vector(-ray.direction.x, ray.direction.y);
} else {
newRayDirection = new Vector(ray.direction.x, -ray.direction.y);
}

let newRay = new Ray(segment[1], newRayDirection, length=remainingLength);
return {
segment: segment,
ray: newRay
};
}


As you have probably guessed, rayToPoints simply calls  splitRay over and over again until the ray hits an input “stopping point”—a guard, the target, or the assassin—or else our finite ray length has been exhausted. The output is a list of points, starting from the original ray’s center, for which adjacent pairs are interpreted as line segments to draw.

  rayToPoints(ray, stoppingPoints) {
let points = [ray.center];
let remainingRay = ray;

while (remainingRay) {
// check if the ray would hit any guards or the target
if (stoppingPoints) {
let hardStops = stoppingPoints.map(p => remainingRay.intersects(p))
.filter(p => p != null);
if (hardStops.length > 0) {
// find first intersection and break
let closestStop = remainingRay.closestToCenter(hardStops);
points.push(closestStop);
break;
}
}

let rayPieces = this.splitRay(remainingRay);
points.push(rayPieces.segment[1]);
remainingRay = rayPieces.ray;
}

return points;
}


That’s sufficient to draw the shot emanating from the assassin. This method is called every time the mouse moves.

## Optimal guards

The function to compute the optimal position of the guards takes as input the containing rectangle, the assassin, and the target, and produces as output a list of 16 points.

/*
* Compute the 16 optimal guards to prevent the assassin from hitting the
* target.
*/
function computeOptimalGuards(square, assassin, target) {
...
}


If you read Tai-Danae’s proof, you’ll know that this construction is to

1. Compute mirrors of the target across the top, the right, and the top+right of the rectangle. Call this resulting thing the 4-mirrored-targets.
2. Replicate the 4-mirrored-targets four times, by translating three of the copies left by the entire width of the 4-mirrored-targets shape, down by the entire height, and both left-and-down.
3. Now you have 16 copies of the target, and one assassin. This gives 16 line segments from assassin-to-target-copy. Place a guard at the midpoint of each of these line segments.
4. Finally, apply the reverse translation and reverse mirroring to return the guards to the original square.

Due to WordPress being a crappy blogging platform I need to migrate off of, the code snippets below have been magically disappearing. I’ve included links to github lines as well.

Step 1 (after adding simple helper functions on Rectangle to do the mirroring):

  // First compute the target copies in the 4 mirrors
let target1 = target.copy();
let target2 = square.mirrorTop(target);
let target3 = square.mirrorRight(target);
let target4 = square.mirrorTop(square.mirrorRight(target));
target1.guardLabel = 1;
target2.guardLabel = 2;
target3.guardLabel = 3;
target4.guardLabel = 4;

  // for each mirrored target, compute the four two-square-length translates
let mirroredTargets = [target1, target2, target3, target4];
let horizontalShift = 2 * square.width();
let verticalShift = 2 * square.height();
let translateLeft = new Vector(-horizontalShift, 0);
let translateRight = new Vector(horizontalShift, 0);
let translateUp = new Vector(0, verticalShift);
let translateDown = new Vector(0, -verticalShift);

let translatedTargets = [];
for (let i = 0; i < mirroredTargets.length; i++) {
let target = mirroredTargets[i];
translatedTargets.push([
target,
]);
}


Step 3, computing the midpoints:

  // compute the midpoints between the assassin and each translate
let translatedMidpoints = [];
for (let i = 0; i  t.midpoint(assassin)));
}


Step 4, returning the guards back to the original square, is harder than it seems, because the midpoint of an assassin-to-target-copy segment might not be in the same copy of the square as the target-copy being fired at. This means you have to detect which square copy the midpoint lands in, and use that to determine which operations are required to invert. This results in the final block of this massive function.

  // determine which of the four possible translates the midpoint is in
// and reverse the translation. Since midpoints can end up in completely
// different copies of the square, we have to check each one for all cases.
function untranslate(point) {
if (point.x  square.bottomLeft.y) {
} else if (point.x >= square.bottomLeft.x && point.y <= square.bottomLeft.y) {
} else if (point.x < square.bottomLeft.x && point.y <= square.bottomLeft.y) {
} else {
return point;
}
}

// undo the translations to get the midpoints back to the original 4-mirrored square.
let untranslatedMidpoints = [];
for (let i = 0; i  square.topRight.x && point.y > square.topRight.y) {
return square.mirrorTop(square.mirrorRight(point));
} else if (point.x > square.topRight.x && point.y <= square.topRight.y) {
return square.mirrorRight(point);
} else if (point.x  square.topRight.y) {
return square.mirrorTop(point);
} else {
return point;
}
}

return untranslatedMidpoints.map(unmirror);


And that’s all there is to it!

## Improvements, if I only had the time

There are a few improvements I’d like to make to this puzzle, but haven’t made the time (I’m writing a book, after all!).

1. Be able to drag the guards around.
2. Create new guards from an empty set of guards, with a button to “reveal” the solution.
3. Include a toggle that, when pressed, darkens the entire region of the square that can be hit by the assassin. For example, this would allow you to see if the target is in the only possible safe spot, or if there are multiple safe spots for a given configuration.
4. Perhaps darken the vulnerable spots by the number of possible paths that hit it, up to some limit.
5. The most complicated one: generalize to an arbitrary polygon (convex or not!), for which there may be no optional solution. The visualization would allow you to look for a solution using 2-4.

Pull requests are welcome if you attempt any of these improvements.

Until next time!

# Binary Search on Graphs

Binary search is one of the most basic algorithms I know. Given a sorted list of comparable items and a target item being sought, binary search looks at the middle of the list, and compares it to the target. If the target is larger, we repeat on the smaller half of the list, and vice versa.

With each comparison the binary search algorithm cuts the search space in half. The result is a guarantee of no more than $\log(n)$ comparisons, for a total runtime of $O(\log n)$. Neat, efficient, useful.

There’s always another angle.

What if we tried to do binary search on a graph? Most graph search algorithms, like breadth- or depth-first search, take linear time, and they were invented by some pretty smart cookies. So if binary search on a graph is going to make any sense, it’ll have to use more information beyond what a normal search algorithm has access to.

For binary search on a list, it’s the fact that the list is sorted, and we can compare against the sought item to guide our search. But really, the key piece of information isn’t related to the comparability of the items. It’s that we can eliminate half of the search space at every step. The “compare against the target” step can be thought of a black box that replies to queries of the form, “Is this the thing I’m looking for?” with responses of the form, “Yes,” or, “No, but look over here instead.”

As long as the answers to your queries are sufficiently helpful, meaning they allow you to cut out large portions of your search space at each step, then you probably have a good algorithm on your hands. Indeed, there’s a natural model for graphs, defined in a 2015 paper of Emamjomeh-Zadeh, Kempe, and Singhal that goes as follows.

You’re given as input an undirected, weighted graph $G = (V,E)$, with weights $w_e$ for $e \in E$. You can see the entire graph, and you may ask questions of the form, “Is vertex $v$ the target?” Responses will be one of two things:

• Yes (you win!)
• No, but $e = (v, w)$ is an edge out of $v$ on a shortest path from $v$ to the true target.

Your goal is to find the target vertex with the minimum number of queries.

Obviously this only works if $G$ is connected, but slight variations of everything in this post work for disconnected graphs. (The same is not true in general for directed graphs)

When the graph is a line, this “reduces” to binary search in the sense that the same basic idea of binary search works: start in the middle of the graph, and the edge you get in response to a query will tell you in which half of the graph to continue.

And if we make this example only slightly more complicated, the generalization should become obvious:

Here, we again start at the “center vertex,” and the response to our query will eliminate one of the two halves. But then how should we pick the next vertex, now that we no longer have a linear order to rely on? It should be clear, choose the “center vertex” of whichever half we end up in. This choice can be formalized into a rule that works even when there’s not such obvious symmetry, and it turns out to always be the right choice.

Definition: median of a weighted graph $G$ with respect to a subset of vertices $S \subset V$ is a vertex $v \in V$ (not necessarily in $S$) which minimizes the sum of distances to vertices in $S$. More formally, it minimizes

$\Phi_S(v) = \sum_{u \in S} d(v, u)$,

where $d(u,v)$ is the sum of the edge weights along a shortest path from $v$ to $u$.

And so generalizing binary search to this query-model on a graph results in the following algorithm, which whittles down the search space by querying the median at every step.

Algorithm: Binary search on graphs. Input is a graph $G = (V,E)$.

• Start with a set of candidates $S = V$.
• While we haven’t found the target and $|S| > 1$:
• Query the median $v$ of $S$, and stop if you’ve found the target.
• Otherwise, let $e = (v, w)$ be the response edge, and compute the set of all vertices $x \in V$ for which $e$ is on a shortest path from $v$ to $x$. Call this set $T$.
• Replace $S$ with $S \cap T$.
• Output the only remaining vertex in $S$

Indeed, as we’ll see momentarily, a python implementation is about as simple. The meat of the work is in computing the median and the set $T$, both of which are slight variants of Dijkstra’s algorithm for computing shortest paths.

The theorem, which is straightforward and well written by Emamjomeh-Zadeh et al. (only about a half page on page 5), is that this algorithm requires only $O(\log(n))$ queries, just like binary search.

Before we dive into an implementation, there’s a catch. Even though we are guaranteed only $\log(n)$ many queries, because of our Dijkstra’s algorithm implementation, we’re definitely not going to get a logarithmic time algorithm. So in what situation would this be useful?

Here’s where we use the “theory” trick of making up a fanciful problem and only later finding applications for it (which, honestly, has been quite successful in computer science). In this scenario we’re treating the query mechanism as a black box. It’s natural to imagine that the queries are expensive, and a resource we want to optimize for. As an example the authors bring up in a followup paper, the graph might be the set of clusterings of a dataset, and the query involves a human looking at the data and responding that a cluster should be split, or that two clusters should be joined. Of course, for clustering the underlying graph is too large to process, so the median-finding algorithm needs to be implicit. But the essential point is clear: sometimes the query is the most expensive part of the algorithm.

Alright, now let’s implement it! The complete code is on Github as always.

## Always be implementing

We start with a slight variation of Dijkstra’s algorithm. Here we’re given as input a single “starting” vertex, and we produce as output a list of all shortest paths from the start to all possible destination vertices.

from collections import defaultdict
from collections import namedtuple

Edge = namedtuple('Edge', ('source', 'target', 'weight'))

class Graph:
# A bare-bones implementation of a weighted, undirected graph
def __init__(self, vertices, edges=tuple()):
self.vertices = vertices
self.incident_edges = defaultdict(list)

for edge in edges:
edge[0],
edge[1],
1 if len(edge) == 2 else edge[2]  # optional weight
)

self.incident_edges[u].append(Edge(u, v, weight))
self.incident_edges[v].append(Edge(v, u, weight))

def edge(self, u, v):
return [e for e in self.incident_edges[u] if e.target == v][0]


And then, since most of the work in Dijkstra’s algorithm is tracking information that you build up as you search the graph, we define the “output” data structure, a dictionary of edge weights paired with back-pointers for the discovered shortest paths.

class DijkstraOutput:
def __init__(self, graph, start):
self.start = start
self.graph = graph

# the smallest distance from the start to the destination v
self.distance_from_start = {v: math.inf for v in graph.vertices}
self.distance_from_start[start] = 0

# a list of predecessor edges for each destination
# to track a list of possibly many shortest paths
self.predecessor_edges = {v: [] for v in graph.vertices}

def found_shorter_path(self, vertex, edge, new_distance):
# update the solution with a newly found shorter path
self.distance_from_start[vertex] = new_distance

if new_distance < self.distance_from_start[vertex]:
self.predecessor_edges[vertex] = [edge]
else:  # tie for multiple shortest paths
self.predecessor_edges[vertex].append(edge)

def path_to_destination_contains_edge(self, destination, edge):
predecessors = self.predecessor_edges[destination]
if edge in predecessors:
return True
return any(self.path_to_destination_contains_edge(e.source, edge)
for e in predecessors)

def sum_of_distances(self, subset=None):
subset = subset or self.graph.vertices
return sum(self.distance_from_start[x] for x in subset)


The actual Dijkstra algorithm then just does a “breadth-first” (priority-queue-guided) search through $G$, updating the metadata as it finds shorter paths.

def single_source_shortest_paths(graph, start):
'''
Compute the shortest paths and distances from the start vertex to all
possible destination vertices. Return an instance of DijkstraOutput.
'''
output = DijkstraOutput(graph, start)
visit_queue = [(0, start)]

while len(visit_queue) > 0:
priority, current = heapq.heappop(visit_queue)

for incident_edge in graph.incident_edges[current]:
v = incident_edge.target
weight = incident_edge.weight
distance_from_current = output.distance_from_start[current] + weight

if distance_from_current <= output.distance_from_start[v]:
output.found_shorter_path(v, incident_edge, distance_from_current)
heapq.heappush(visit_queue, (distance_from_current, v))

return output


Finally, we implement the median-finding and $T$-computing subroutines:

def possible_targets(graph, start, edge):
'''
Given an undirected graph G = (V,E), an input vertex v in V, and an edge e
incident to v, compute the set of vertices w such that e is on a shortest path from
v to w.
'''
dijkstra_output = dijkstra.single_source_shortest_paths(graph, start)
return set(v for v in graph.vertices
if dijkstra_output.path_to_destination_contains_edge(v, edge))

def find_median(graph, vertices):
'''
Compute as output a vertex in the input graph which minimizes the sum of distances
to the input set of vertices
'''
best_dijkstra_run = min(
(single_source_shortest_paths(graph, v) for v in graph.vertices),
key=lambda run: run.sum_of_distances(vertices)
)
return best_dijkstra_run.start


And then the core algorithm

QueryResult = namedtuple('QueryResult', ('found_target', 'feedback_edge'))

def binary_search(graph, query):
'''
Find a target node in a graph, with queries of the form "Is x the target?"
and responses either "You found the target!" or "Here is an edge on a shortest
path to the target."
'''
candidate_nodes = set(x for x in graph.vertices)  # copy

while len(candidate_nodes) > 1:
median = find_median(graph, candidate_nodes)
query_result = query(median)

if query_result.found_target:
return median
else:
edge = query_result.feedback_edge
legal_targets = possible_targets(graph, median, edge)
candidate_nodes = candidate_nodes.intersection(legal_targets)

return candidate_nodes.pop()


Here’s an example of running it on the example graph we used earlier in the post:

'''
Graph looks like this tree, with uniform weights

a       k
b     j
cfghi
d     l
e       m
'''
G = Graph(['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i',
'j', 'k', 'l', 'm'],
[
('a', 'b'),
('b', 'c'),
('c', 'd'),
('d', 'e'),
('c', 'f'),
('f', 'g'),
('g', 'h'),
('h', 'i'),
('i', 'j'),
('j', 'k'),
('i', 'l'),
('l', 'm'),
])

def simple_query(v):
ans = input("is '%s' the target? [y/N] " % v)
if ans and ans.lower()[0] == 'y':
return QueryResult(True, None)
else:
print("Please input a vertex on the shortest path between"
" '%s' and the target. The graph is: " % v)
for w in G.incident_edges:
print("%s: %s" % (w, G.incident_edges[w]))

target = None
while target not in G.vertices:
target = input("Input neighboring vertex of '%s': " % v)

return QueryResult(
False,
G.edge(v, target)
)

output = binary_search(G, simple_query)
print("Found target: %s" % output)


The query function just prints out a reminder of the graph and asks the user to answer the query with a yes/no and a relevant edge if the answer is no.

An example run:

is 'g' the target? [y/N] n
Please input a vertex on the shortest path between 'g' and the target. The graph is:
e: [Edge(source='e', target='d', weight=1)]
i: [Edge(source='i', target='h', weight=1), Edge(source='i', target='j', weight=1), Edge(source='i', target='l', weight=1)]
g: [Edge(source='g', target='f', weight=1), Edge(source='g', target='h', weight=1)]
l: [Edge(source='l', target='i', weight=1), Edge(source='l', target='m', weight=1)]
k: [Edge(source='k', target='j', weight=1)]
j: [Edge(source='j', target='i', weight=1), Edge(source='j', target='k', weight=1)]
c: [Edge(source='c', target='b', weight=1), Edge(source='c', target='d', weight=1), Edge(source='c', target='f', weight=1)]
f: [Edge(source='f', target='c', weight=1), Edge(source='f', target='g', weight=1)]
m: [Edge(source='m', target='l', weight=1)]
d: [Edge(source='d', target='c', weight=1), Edge(source='d', target='e', weight=1)]
h: [Edge(source='h', target='g', weight=1), Edge(source='h', target='i', weight=1)]
b: [Edge(source='b', target='a', weight=1), Edge(source='b', target='c', weight=1)]
a: [Edge(source='a', target='b', weight=1)]
Input neighboring vertex of 'g': f
is 'c' the target? [y/N] n
Please input a vertex on the shortest path between 'c' and the target. The graph is:
[...]
Input neighboring vertex of 'c': d
is 'd' the target? [y/N] n
Please input a vertex on the shortest path between 'd' and the target. The graph is:
[...]
Input neighboring vertex of 'd': e
Found target: e


## A likely story

The binary search we implemented in this post is pretty minimal. In fact, the more interesting part of the work of Emamjomeh-Zadeh et al. is the part where the response to the query can be wrong with some unknown probability.

In this case, there can be many shortest paths that are valid responses to a query, in addition to all the invalid responses. In particular, this rules out the strategy of asking the same query multiple times and taking the majority response. If the error rate is 1/3, and there are two shortest paths to the target, you can get into a situation in which you see three responses equally often and can’t choose which one is the liar.

Instead, the technique Emamjomeh-Zadeh et al. use is based on the Multiplicative Weights Update Algorithm (it strikes again!). Each query gives a multiplicative increase (or decrease) on the set of nodes that are consistent targets under the assumption that query response is correct. There are a few extra details and some postprocessing to avoid unlikely outcomes, but that’s the basic idea. Implementing it would be an excellent exercise for readers interested in diving deeper into a recent research paper (or to flex their math muscles).

But even deeper, this model of “query and get advice on how to improve” is a classic  learning model first formally studied by Dana Angluin (my academic grand-advisor). In her model, one wants to design an algorithm to learn a classifier. The allowed queries are membership and equivalence queries. A membership is essentially, “What’s its label of this element?” and an equivalence query has the form, “Is this the right classifier?” If the answer is no, a mislabeled example is provided.

This is different from the usual machine learning assumption, because the learning algorithm gets to construct an example it wants to get more information about, instead of simply relying on a randomly generated subset of data. The goal is to minimize the number of queries before the target hypothesis is learned exactly. And indeed, as we saw in this post, if you have a little extra time to analyze the problem space, you can craft queries that extract quite a lot of information.

Indeed, the model we presented here for binary search on graphs is the natural analogue of an equivalence query for a search problem: instead of a mislabeled counterexample, you get a nudge in the right direction toward the target. Pretty neat!

There are a few directions we could take from here: (1) implement the Multiplicative Weights version of the algorithm, (2) apply this technique to a problem like ranking or clustering, or (3) cover theoretical learning models like membership and equivalence queries in more detail. What interests you?

Until next time!

# Formulating the Support Vector Machine Optimization Problem

## The hypothesis and the setup

This blog post has an interactive demo (mostly used toward the end of the post). The source for this demo is available in a Github repository.

Last time we saw how the inner product of two vectors gives rise to a decision rule: if $w$ is the normal to a line (or hyperplane) $L$, the sign of the inner product $\langle x, w \rangle$ tells you whether $x$ is on the same side of $L$ as $w$.

Let’s translate this to the parlance of machine-learning. Let $x \in \mathbb{R}^n$ be a training data point, and $y \in \{ 1, -1 \}$ is its label (green and red, in the images in this post). Suppose you want to find a hyperplane which separates all the points with -1 labels from those with +1 labels (assume for the moment that this is possible). For this and all examples in this post, we’ll use data in two dimensions, but the math will apply to any dimension.

Some data labeled red and green, which is separable by a hyperplane (line).

The hypothesis we’re proposing to separate these points is a hyperplane, i.e. a linear subspace that splits all of $\mathbb{R}^n$ into two halves. The data that represents this hyperplane is a single vector $w$, the normal to the hyperplane, so that the hyperplane is defined by the solutions to the equation $\langle x, w \rangle = 0$.

As we saw last time, $w$ encodes the following rule for deciding if a new point $z$ has a positive or negative label.

$\displaystyle h_w(z) = \textup{sign}(\langle w, x \rangle)$

You’ll notice that this formula only works for the normals $w$ of hyperplanes that pass through the origin, and generally we want to work with data that can be shifted elsewhere. We can resolve this by either adding a fixed term $b \in \mathbb{R}$—often called a bias because statisticians came up with it—so that the shifted hyperplane is the set of solutions to $\langle x, w \rangle + b = 0$. The shifted decision rule is:

$\displaystyle h_w(z) = \textup{sign}(\langle w, x \rangle + b)$

Now the hypothesis is the pair of vector-and-scalar $w, b$.

The key intuitive idea behind the formulation of the SVM problem is that there are many possible separating hyperplanes for a given set of labeled training data. For example, here is a gif showing infinitely many choices.

The question is: how can we find the separating hyperplane that not only separates the training data, but generalizes as well as possible to new data? The assumption of the SVM is that a hyperplane which separates the points, but is also as far away from any training point as possible, will generalize best.

While contrived, it’s easy to see that the separating hyperplane is as far as possible from any training point.

More specifically, fix a labeled dataset of points $(x_i, y_i)$, or more precisely:

$\displaystyle D = \{ (x_i, y_i) \mid i = 1, \dots, m, x_i \in \mathbb{R}^{n}, y_i \in \{1, -1\} \}$

And a hypothesis defined by the normal $w \in \mathbb{R}^{n}$ and a shift $b \in \mathbb{R}$. Let’s also suppose that $(w,b)$ defines a hyperplane that correctly separates all the training data into the two labeled classes, and we just want to measure its quality. That measure of quality is the length of its margin.

Definition: The geometric margin of a hyperplane $w$ with respect to a dataset $D$ is the shortest distance from a training point $x_i$ to the hyperplane defined by $w$.

The best hyperplane has the largest possible margin.

This margin can even be computed quite easily using our work from last post. The distance from $x$ to the hyperplane defined by $w$ is the same as the length of the projection of $x$ onto $w$. And this is just computed by an inner product.

If the tip of the $x$ arrow is the point in question, then $a$ is the dot product, and $b$ the distance from $x$ to the hyperplane $L$ defined by $w$.

## A naive optimization objective

If we wanted to, we could stop now and define an optimization problem that would be very hard to solve. It would look like this:

\displaystyle \begin{aligned} & \max_{w} \min_{x_i} \left | \left \langle x_i, \frac{w}{\|w\|} \right \rangle + b \right | & \\ \textup{subject to \ \ } & \textup{sign}(\langle x_i, w \rangle + b) = \textup{sign}(y_i) & \textup{ for every } i = 1, \dots, m \end{aligned}

The formulation is hard. The reason is it’s horrifyingly nonlinear. In more detail:

1. The constraints are nonlinear due to the sign comparisons.
2. There’s a min and a max! A priori, we have to do this because we don’t know which point is going to be the closest to the hyperplane.
3. The objective is nonlinear in two ways: the absolute value and the projection requires you to take a norm and divide.

The rest of this post (and indeed, a lot of the work in grokking SVMs) is dedicated to converting this optimization problem to one in which the constraints are all linear inequalities and the objective is a single, quadratic polynomial we want to minimize or maximize.

Along the way, we’ll notice some neat features of the SVM.

## Trick 1: linearizing the constraints

To solve the first problem, we can use a trick. We want to know whether $\textup{sign}(\langle x_i, w \rangle + b) = \textup{sign}(y_i)$ for a labeled training point $(x_i, y_i)$. The trick is to multiply them together. If their signs agree, then their product will be positive, otherwise it will be negative.

So each constraint becomes:

$\displaystyle (\langle x_i, w \rangle + b) \cdot y_i \geq 0$

This is still linear because $y_i$ is a constant (input) to the optimization problem. The variables are the coefficients of $w$.

The left hand side of this inequality is often called the functional margin of a training point, since, as we will see, it still works to classify $x_i$, even if $w$ is scaled so that it is no longer a unit vector. Indeed, the sign of the inner product is independent of how $w$ is scaled.

## Trick 1.5: the optimal solution is midway between classes

This small trick is to notice that if $w$ is the supposed optimal separating hyperplane, i.e. its margin is maximized, then it must necessarily be exactly halfway in between the closest points in the positive and negative classes.

In other words, if $x_+$ and $x_-$ are the closest points in the positive and negative classes, respectively, then $\langle x_{+}, w \rangle + b = -(\langle x_{-}, w \rangle + b)$. If this were not the case, then you could adjust the bias, shifting the decision boundary along $w$ until it they are exactly equal, and you will have increased the margin. The closest point, say $x_+$ will have gotten farther away, and the closest point in the opposite class, $x_-$ will have gotten closer, but will not be closer than $x_+$.

## Trick 2: getting rid of the max + min

Resolving this problem essentially uses the fact that the hypothesis, which comes in the form of the normal vector $w$, has a degree of freedom in its length. To explain the details of this trick, we’ll set $b=0$ which simplifies the intuition.

Indeed, in the animation below, I can increase or decrease the length of $w$ without changing the decision boundary.

I have to keep my hand very steady (because I was too lazy to program it so that it only increases/decreases in length), but you can see the point. The line is perpendicular to the normal vector, and it doesn’t depend on the length.

Let’s combine this with tricks 1 and 1.5. If we increase the length of $w$, that means the absolute values of the dot products $\langle x_i, w \rangle$ used in the constraints will all increase by the same amount (without changing their sign). Indeed, for any vector $a$ we have $\langle a, w \rangle = \|w \| \cdot \langle a, w / \| w \| \rangle$.

In this world, the inner product measurement of distance from a point to the hyperplane is no longer faithful. The true distance is $\langle a, w / \| w \| \rangle$, but the distance measured by $\langle a, w \rangle$ is measured in units of $1 / \| w \|$.

In this example, the two numbers next to the green dot represent the true distance of the point from the hyperplane, and the dot product of the point with the normal (respectively). The dashed lines are the solutions to <x, w> = 1. The magnitude of w is 2.2, the inverse of that is 0.46, and indeed 2.2 = 4.8 * 0.46 (we’ve rounded the numbers).

Now suppose we had the optimal hyperplane and its normal $w$. No matter how near (or far) the nearest positively labeled training point $x$ is, we could scale the length of $w$ to force $\langle x, w \rangle = 1$. This is the core of the trick. One consequence is that the actual distance from $x$ to the hyperplane is $\frac{1}{\| w \|} = \langle x, w / \| w \| \rangle$.

The same as above, but with the roles reversed. We’re forcing the inner product of the point with w to be 1. The true distance is unchanged.

In particular, if we force the closest point to have inner product 1, then all other points will have inner product at least 1. This has two consequences. First, our constraints change to $\langle x_i, w \rangle \cdot y_i \geq 1$ instead of $\geq 0$. Second, we no longer need to ask which point is closest to the candidate hyperplane! Because after all, we never cared which point it was, just how far away that closest point was. And now we know that it’s exactly $1 / \| w \|$ away. Indeed, if the optimal points weren’t at that distance, then that means the closest point doesn’t exactly meet the constraint, i.e. that $\langle x, w \rangle > 1$ for every training point $x$. We could then scale $w$ shorter until $\langle x, w \rangle = 1$, hence increasing the margin $1 / \| w \|$.

In other words, the coup de grâce, provided all the constraints are satisfied, the optimization objective is just to maximize $1 / \| w \|$, a.k.a. to minimize $\| w \|$.

This intuition is clear from the following demonstration, which you can try for yourself. In it I have a bunch of positively and negatively labeled points, and the line in the center is the candidate hyperplane with normal $w$ that you can drag around. Each training point has two numbers next to it. The first is the true distance from that point to the candidate hyperplane; the second is the inner product with $w$. The two blue dashed lines are the solutions to $\langle x, w \rangle = \pm 1$. To solve the SVM by hand, you have to ensure the second number is at least 1 for all green points, at most -1 for all red points, and then you have to make $w$ as short as possible. As we’ve discussed, shrinking $w$ moves the blue lines farther away from the separator, but in order to satisfy the constraints the blue lines can’t go further than any training point. Indeed, the optimum will have those blue lines touching a training point on each side.

I bet you enjoyed watching me struggle to solve it. And while it’s probably not the optimal solution, the idea should be clear.

The final note is that, since we are now minimizing $\| w \|$, a formula which includes a square root, we may as well minimize its square $\| w \|^2 = \sum_j w_j^2$. We will also multiply the objective by $1/2$, because when we eventually analyze this problem we will take a derivative, and the square in the exponent and the $1/2$ will cancel.

## The final form of the problem

Our optimization problem is now the following (including the bias again):

\displaystyle \begin{aligned} & \min_{w} \frac{1}{2} \| w \|^2 & \\ \textup{subject to \ \ } & (\langle x_i, w \rangle + b) \cdot y_i \geq 1 & \textup{ for every } i = 1, \dots, m \end{aligned}

This is much simpler to analyze. The constraints are all linear inequalities (which, because of linear programming, we know are tractable to optimize). The objective to minimize, however, is a convex quadratic function of the input variables—a sum of squares of the inputs.

Such problems are generally called quadratic programming problems (or QPs, for short). There are general methods to find solutions! However, they often suffer from numerical stability issues and have less-than-satisfactory runtime. Luckily, the form in which we’ve expressed the support vector machine problem is specific enough that we can analyze it directly, and find a way to solve it without appealing to general-purpose numerical solvers.

We will tackle this problem in a future post (planned for two posts sequel to this one). Before we close, let’s just make a few more observations about the solution to the optimization problem.

## Support Vectors

In Trick 1.5 we saw that the optimal separating hyperplane has to be exactly halfway between the two closest points of opposite classes. Moreover, we noticed that, provided we’ve scaled $\| w \|$ properly, these closest points (there may be multiple for positive and negative labels) have to be exactly “distance” 1 away from the separating hyperplane.

Another way to phrase this without putting “distance” in scare quotes is to say that, if $w$ is the normal vector of the optimal separating hyperplane, the closest points lie on the two lines $\langle x_i, w \rangle + b = \pm 1$.

Now that we have some intuition for the formulation of this problem, it isn’t a stretch to realize the following. While a dataset may include many points from either class on these two lines $\langle x_i, w \rangle = \pm 1$, the optimal hyperplane itself does not depend on any of the other points except these closest points.

This fact is enough to give these closest points a special name: the support vectors.

We’ll actually prove that support vectors “are all you need” with full rigor and detail next time, when we cast the optimization problem in this post into the “dual” setting. To avoid vague names, the formulation described in this post called the “primal” problem. The dual problem is derived from the primal problem, with special variables and constraints chosen based on the primal variables and constraints. Next time we’ll describe in brief detail what the dual does and why it’s important, but we won’t have nearly enough time to give a full understanding of duality in optimization (such a treatment would fill a book).

When we compute the dual of the SVM problem, we will see explicitly that the hyperplane can be written as a linear combination of the support vectors. As such, once you’ve found the optimal hyperplane, you can compress the training set into just the support vectors, and reproducing the same optimal solution becomes much, much faster. You can also use the support vectors to augment the SVM to incorporate streaming data (throw out all non-support vectors after every retraining).

Eventually, when we get to implementing the SVM from scratch, we’ll see all this in action.

Until then!

# The Inner Product as a Decision Rule

The standard inner product of two vectors has some nice geometric properties. Given two vectors $x, y \in \mathbb{R}^n$, where by $x_i$ I mean the $i$-th coordinate of $x$, the standard inner product (which I will interchangeably call the dot product) is defined by the formula

$\displaystyle \langle x, y \rangle = x_1 y_1 + \dots + x_n y_n$

This formula, simple as it is, produces a lot of interesting geometry. An important such property, one which is discussed in machine learning circles more than pure math, is that it is a very convenient decision rule.

In particular, say we’re in the Euclidean plane, and we have a line $L$ passing through the origin, with $w$ being a unit vector perpendicular to $L$ (“the normal” to the line).

If you take any vector $x$, then the dot product $\langle x, w \rangle$ is positive if $x$ is on the same side of $L$ as $w$, and negative otherwise. The dot product is zero if and only if $x$ is exactly on the line $L$, including when $x$ is the zero vector.

Left: the dot product of $w$ and $x$ is positive, meaning they are on the same side of $w$. Right: The dot product is negative, and they are on opposite sides.

Here is an interactive demonstration of this property. Click the image below to go to the demo, and you can drag the vector arrowheads and see the decision rule change.

Click above to go to the demo

The code for this demo is available in a github repository.

It’s always curious, at first, that multiplying and summing produces such geometry. Why should this seemingly trivial arithmetic do anything useful at all?

The core fact that makes it work, however, is that the dot product tells you how one vector projects onto another. When I say “projecting” a vector $x$ onto another vector $w$, I mean you take only the components of $x$ that point in the direction of $w$. The demo shows what the result looks like using the red (or green) vector.

In two dimensions this is easy to see, as you can draw the triangle which has $x$ as the hypotenuse, with $w$ spanning one of the two legs of the triangle as follows:

If we call $a$ the (vector) leg of the triangle parallel to $w$, while $b$ is the dotted line (as a vector, parallel to $L$), then as vectors $x = a + b$. The projection of $x$ onto $w$ is just $a$.

Another way to think of this is that the projection is $x$, modified by removing any part of $x$ that is perpendicular to $w$. Using some colorful language: you put your hands on either side of $x$ and $w$, and then you squish $x$ onto $w$ along the line perpendicular to $w$ (i.e., along $b$).

And if $w$ is a unit vector, then the length of $a$—that is, the length of the projection of $x$ onto $w$—is exactly the inner product product $\langle x, w \rangle$.

Moreover, if the angle between $x$ and $w$ is larger than 90 degrees, the projected vector will point in the opposite direction of $w$, so it’s really a “signed” length.

Left: the projection points in the same direction as $w$. Right: the projection points in the opposite direction.

And this is precisely why the decision rule works. This 90-degree boundary is the line perpendicular to $w$.

More technically said: Let $x, y \in \mathbb{R}^n$ be two vectors, and $\langle x,y \rangle$ their dot product. Define by $\| y \|$ the length of $y$, specifically $\sqrt{\langle y, y \rangle}$. Define by $\text{proj}_{y}(x)$ by first letting $y' = \frac{y}{\| y \|}$, and then let $\text{proj}_{y}(x) = \langle x,y' \rangle y'$. In words, you scale $y$ to a unit vector $y'$, use the result to compute the inner product, and then scale $y$ so that it’s length is $\langle x, y' \rangle$. Then

Theorem: Geometrically, $\text{proj}_y(x)$ is the projection of $x$ onto the line spanned by $y$.

This theorem is true for any $n$-dimensional vector space, since if you have two vectors you can simply apply the reasoning for 2-dimensions to the 2-dimensional plane containing $x$ and $y$. In that case, the decision boundary for a positive/negative output is the entire $n-1$ dimensional hyperplane perpendicular to $y$ (the projected vector).

In fact, the usual formula for the angle between two vectors, i.e. the formula $\langle x, y \rangle = \|x \| \cdot \| y \| \cos \theta$, is a restatement of the projection theorem in terms of trigonometry. The $\langle x, y' \rangle$ part of the projection formula (how much you scale the output) is equal to $\| x \| \cos \theta$. At the end of this post we have a proof of the cosine-angle formula above.

Part of why this decision rule property is so important is that this is a linear function, and linear functions can be optimized relatively easily. When I say that, I specifically mean that there are many known algorithms for optimizing linear functions, which don’t have obscene runtime or space requirements. This is a big reason why mathematicians and statisticians start the mathematical modeling process with linear functions. They’re inherently simpler.

In fact, there are many techniques in machine learning—a prominent one is the so-called Kernel Trick—that exist solely to take data that is not inherently linear in nature (cannot be fruitfully analyzed by linear methods) and transform it into a dataset that is. Using the Kernel Trick as an example to foreshadow some future posts on Support Vector Machines, the idea is to take data which cannot be separated by a line, and transform it (usually by adding new coordinates) so that it can. Then the decision rule, computed in the larger space, is just a dot product. Irene Papakonstantinou neatly demonstrates this with paper folding and scissors. The tradeoff is that the size of the ambient space increases, and it might increase so much that it makes computation intractable. Luckily, the Kernel Trick avoids this by remembering where the data came from, so that one can take advantage of the smaller space to compute what would be the inner product in the larger space.

Next time we’ll see how this decision rule shows up in an optimization problem: finding the “best” hyperplane that separates an input set of red and blue points into monochromatic regions (provided that is possible). Finding this separator is core subroutine of the Support Vector Machine technique, and therein lie interesting algorithms. After we see the core SVM algorithm, we’ll see how the Kernel Trick fits into the method to allow nonlinear decision boundaries.

Proof of the cosine angle formula

Theorem: The inner product $\langle v, w \rangle$ is equal to $\| v \| \| w \| \cos(\theta)$, where $\theta$ is the angle between the two vectors.

Note that this angle is computed in the 2-dimensional subspace spanned by $v, w$, viewed as a typical flat plane, and this is a 2-dimensional plane regardless of the dimension of $v, w$.

Proof. If either $v$ or $w$ is zero, then both sides of the equation are zero and the theorem is trivial, so we may assume both are nonzero. Label a triangle with sides $v,w$ and the third side $v-w$. Now the length of each side is $\| v \|, \| w\|,$ and $\| v-w \|$, respectively. Assume for the moment that $\theta$ is not 0 or 180 degrees, so that this triangle is not degenerate.

The law of cosines allows us to write

$\displaystyle \| v - w \|^2 = \| v \|^2 + \| w \|^2 - 2 \| v \| \| w \| \cos(\theta)$

Moreover, The left hand side is the inner product of $v-w$ with itself, i.e. $\| v - w \|^2 = \langle v-w , v-w \rangle$. We’ll expand $\langle v-w, v-w \rangle$ using two facts. The first is trivial from the formula, that inner product is symmetric: $\langle v,w \rangle = \langle w, v \rangle$. Second is that the inner product is linear in each input. In particular for the first input: $\langle x + y, z \rangle = \langle x, z \rangle + \langle y, z \rangle$ and $\langle cx, z \rangle = c \langle x, z \rangle$. The same holds for the second input by symmetry of the two inputs. Hence we can split up $\langle v-w, v-w \rangle$ as follows.

\displaystyle \begin{aligned} \langle v-w, v-w \rangle &= \langle v, v-w \rangle - \langle w, v-w \rangle \\ &= \langle v, v \rangle - \langle v, w \rangle - \langle w, v \rangle + \langle w, w \rangle \\ &= \| v \|^2 - 2 \langle v, w \rangle + \| w \|^2 \\ \end{aligned}

Combining our two offset equations, we can subtract $\| v \|^2 + \| w \|^2$ from each side and get

$\displaystyle -2 \|v \| \|w \| \cos(\theta) = -2 \langle v, w \rangle,$

Which, after dividing by $-2$, proves the theorem if $\theta \not \in \{0, 180 \}$.

Now if $\theta = 0$ or 180 degrees, the vectors are parallel, so we can write one as a scalar multiple of the other. Say $w = cv$ for $c \in \mathbb{R}$. In that case, $\langle v, cv \rangle = c \| v \| \| v \|$. Now $\| w \| = | c | \| v \|$, since a norm is a length and is hence non-negative (but $c$ can be negative). Indeed, if $v, w$ are parallel but pointing in opposite directions, then $c < 0$, so $\cos(\theta) = -1$, and $c \| v \| = - \| w \|$. Otherwise $c > 0$ and $\cos(\theta) = 1$. This allows us to write $c \| v \| \| v \| = \| w \| \| v \| \cos(\theta)$, and this completes the final case of the theorem.

$\square$