## 4thJLTi Code Jam – Jun 2017

Input:

1 USD = 1.380 SGD

1 SGD = 3.080 MYR

1 MYR = 15.120 INR

1 INR = 0.012 GBP

1 GBP = 1.30 USD

Explanation: Now that we realize, we have to wait substantial amount of time to make any meaningful gain from stock market, we change our focus to money exchange. I am particularly very excited after collecting the above exchange rates using Google search today at 8th Jun 2017. If I start with 10, 000 USD, then convert them to MYR, then convert them to INR and then to GBP and then back to USD, I realize I will end up with 10, 025.50 USD, making USD 25.50 on the same spot within minutes.

Output: USD -> SGD -> MYR -> INR -> GBP -> USD

The next step that I am going to take is to get a list of all money exchanges available to me, somehow collect their exchange rates daily, and push it to a program that will tell me when it sees there is a chance to make some profit.

I might not always be lucky. To check whether my adrenaline rush is justified I looked at the same rates in an online money exchange. The rates from it look like below:

Input:

1 USD = 1.38295 SGD

1 SGD = 3.08614 MYR

1 MYR = 15.0996 INR

1 INR = 0.0119755 GBP

1 GBP = 1.295 USD

As you realize, we will end up losing money.

Output: No luck here

Task: I also realize, given a few hundred currencies and thousands of exchange rates among them, there is a possibility of having a number of ways we can make money. For example, given a set of rates (not as shown in the above example), we could make USD 10, starting with USD 10, 000, using this route: USD -> SGD -> MYR -> USD. Again, we could possibly make SGD 50, starting with SGD 10, 000, using another route: SGD -> MYR -> INR -> SGD. I am happy with just one such route, not necessarily the one providing the most profit.

Index

## 20thFriday Fun Session – 2nd Jun 2017

Given the stock prices for a number of days, in order, we have to buy one stock at one day and then sell it on a later day to maximize the profit.

This is the solution to JLTi Code Jam – May 2017  problem.

### Let us walk through an example

Suppose, we have 8 days’ stock prices, starting at day 1, in order and they are 2, 10, 5, 12, 1, 3, 11, and 9 respectively. We can clearly see, if we buy a stock at day 1, at a price of 2 and then sell it on day 4, at a price of 12; we can make a profit of 10. We could make the same profit had we bought it on day 5, at a price of 1 and then sold it on day 7, at a price of 11. Since we want to make the most profit only once, we would choose say, the first one.

### Accumulator pattern

We could simply run two loops, consider all possible buys and sells, calculate all possible profits (like buy on day 1 and sell it on day 2 with profit 8, buy on day 1 and sell it on day 3 with profit 3, and so on) and find the maximum profit. The below code using accumulator pattern  – traverse a sequence and accumulate a value (here it is maximum), as it goes, can be used to do so.

```void StockN2(double price[], int n, double &maxProfit, int &buyDay, int &sellDay)
{
for(int i=0; i<n; i++)
for(int j=i+1; j<n-1; j++)     {       if(price[j] - price[i] > maxProfit)
{
maxProfit = price[j] - price[i];
sellDay = j;
}
}
}```

For the outer loop that runs (n-1) times the inner loop runs (n-1) times resulting in O(n2), also known as quadratic complexity. Can we do better?

### Divide and conquer solution

When we have an algorithm with O(n2) complexity, and we think about optimizing it, we ask ourselves – what could be better than this. And then O(n log n) comes to our mind. Even though there are many other complexities in between these two, usually O(n log n) is next best after O(n2). And when we see log n, divide and concur comes to our mind.

If we use divide and conquer, we have to divide the input into two sets, find some kind of solutions from both sides and then combine them. What decision can be found from two sides? We can get the maximum profit from each of the two sides. For example, suppose, we have only 4 days’ stock prices: 2, 10, 5 and 12. If we divide them into two, we would get left side: 2 and 10. Right side: 5 and 12. Left side profit would be 8 and the same for right side would be 7. The best between the two would be 8.

However, we can clearly see that the maximum profit is not 8, but 10 when the buy happens in left and sell happens in the right side. It is understandable that local solutions from left and right sides alone would not result in a global optimal solution. Somehow we have to compute a third profit combining the two sides. It is also obvious that buy happens in the left side and sell happens in the right side (after all, we cannot sell before we buy). So we see that the merge phase of the divide and conquer should consider the below 3 profits and find the best among them.

• Maximum profit from left side
• Maximum profit from right side
• Profit by buying at the lowest from left and then selling at the highest in right

The below code is doing this. By the way, we are not tracking the buy day and sell day to keep the focus on the main points. Buy day and sell day can be easily accommodated.

```void StockDivideAndConquerNlogN(double price[], int start, int end, double &maxProfit)
{
if(start == end)
{
// just one value, return
maxProfit = 0;
return;
}

int mid = start + (end-start)/2;

double leftMaxProfit;
StockDivideAndConquerNlogN(price, start, mid, leftMaxProfit);

double rightMaxProfit;
StockDivideAndConquerNlogN(price, mid+1, end, rightMaxProfit);

double minLeft = GetMin(price, start, end);
double maxRight = GetMax(price, start, end);

double minValue = GetMin(price, start, mid);
double maxValue = GetMax(price, mid+1, end);
maxProfit = maxOutOfThree(leftMaxProfit, rightMaxProfit, maxValue - minValue);
}```

For our working example, with 8 days’ stock prices, it looks like below:

The value inside the circle indicates the profit. It is coming from the best of the three as detailed earlier.

#### Computing complexity

How much is the cost? To compute it, we have to find two things – how many levels and how much work is done at each level.

How many levels? We have 8 items. At each level, it is halved. 8 -> 4 -> 2 -> 1. Suppose, in general, we are halving it k times. That means, n is divided by 2, k times. That means, n is divided by 2k and then it becomes 1. 1 = n/2k => n = 2k => log n = k log 2. Since, log 2 = 1, base being 2, k = log n. For n = 8, k = 3.

Level starts at 0 (the root) and ends at k. Hence, the actual number of levels is k+1. For simplicity, the rest of the post would consider the total number of levels to be log n, ignoring the small constant issue of 1.

Next thing to find: how much work is done at each level. We see, at each level, minimum is found from left, costing n/2 and maximum is found from right, also costing n/2. At each level, merging them (computing the third profit and then finding the best among the three) costs iterating n items and a few constant comparisons. Note that, at each level, all n items are present. It is the number of total processed items from all the sub-problems present in that level, not necessarily just from two sub-problems. So for log n levels the total cost is (n * log n) = n log n.

This calculation is explained/performed using master theorem.

### Optimized divide and conquer solution

At each level, we are running loops and iterating n items to get the minimum from left and maximum from right. This loop is not necessary. The minimum and maximum can be computed bottom up doing a constant number of comparisons, at each level.

The below code shows this optimized version:

```void StockDivideAndConquerOptimizedN(double price[], int start, int end, double &maxProfit, double &minValue, double &maxValue)
{
if(start == end)
{
// just one value, return
maxProfit = 0;
minValue = maxValue = price[end];
return;
}

int mid = start + (end-start)/2;

double leftMaxProfit, leftMinValue, leftMaxValue;
StockDivideAndConquerOptimizedN(price, start, mid, leftMaxProfit, leftMinValue, leftMaxValue);

double rightMaxProfit, rightMinValue, rightMaxValue;
StockDivideAndConquerOptimizedN(price, mid + 1, end, rightMaxProfit, rightMinValue, rightMaxValue);

maxProfit = maxOutOfThree(leftMaxProfit, rightMaxProfit, rightMaxValue - leftMinValue);

minValue = leftMinValue > rightMinValue ? rightMinValue : leftMinValue;
maxValue = leftMaxValue > rightMaxValue ? leftMaxValue : rightMaxValue;
}```

#### Computing complexity for this optimized version

In this optimized version, for each of the log n levels, we are still doing some processing. It is no longer n items. Rather, the number of items is decreasing by half at each level, upwards. At the bottom-most level there are n items. One level up, it is reduced by half (due to the merging), shrinking to n/2 items. As it goes up, it gets reduced and at the topmost level it becomes only one item. Let us add the items from each level that we we processing.

n + n/2 + n/22 + n/2+ n/24 + .  .  .  . + n/2k)

=> n (1 + 1/2 + 1/4 + 1/8 + . . .)

=> n * 2 (the convergent series gives 2)

=> 2 n

By discarding the constant terms, we get a complexity of O(n), meaning we can get the maximum profit in linear time.

GitHub: Stock Profit

Index

## 2ndFriday Fun Session – 13th Jan 2017

### What is Fibonacci number?

0, 1, 2, 3, 5, 8 . . . is Fibonacci series where 1st number is 0, 2nd number 1, 3rd number 2 and so on, each number being the sum of its two predecessors.

### What are we talking about here?

We will see that nth Fibonacci number can be found using both recursive and iterative methods. Recursive one will be prohibitively expensive while iterative one will be much more efficient.

### Recursive solution

We can use the following recursive function to get nth Fibonacci number.

```int FibonacciExponential(int n, int &opCounter)
{
opCounter++;

if(n == 0 || n == 1)
return n;

return FibonacciExponential(n-1, opCounter) +
FibonacciExponential(n-2, opCounter);
}```

We have passed an extra parameter to count the number of times the recursive function gets called. For example, for n = 4, we see we have the following calls, each node showing the value of n, with which the function is called. Many a times, for a certain value of n, the function is called numerous times.

We see, if we increase n by one, the tree also expands by one more level. For n = 4 we end up with 24 (2 raised to the 4th power) calls. It will be few less as the right sub-tree is one level less than the left, but they are negligible. When the input number n (input size) goes as an exponent, we call the complexity – exponential. Especially, when we talk about Big O notation, we express it using the upper asymptotic bound. The complexity here is O(2n).

The value 2n when n = 100, is 2100 = 1267650600228229401496703205376. What if each operation takes a millisecond? The execution time would be trillions of years ~ just for n = 100!

### Iterative solution

We could as well run a simple loop by retaining the previous two values and add them to find the present Fibonacci number. For n = 100, we just needed to loop maximum 100 times, requiring 100 operations. That means, we could call the complexity linear and in terms of big O notation it would be O(n).

The below function finds nth Fibonacci number iteratively, in linear time.

```int FibonacciLinear(int n, int &opCounter)
{
opCounter++;

if(n == 0 || n == 1)
return n;

int result = 0;
int previousPrevious = 0;
int immediatePrevious = 1;

for(int i=2; i<=n; i++)
{
opCounter++;

result = immediatePrevious + previousPrevious;

previousPrevious = immediatePrevious;
immediatePrevious = result;
}

return result;;
}```

Considering each operation taking 1 millisecond, we are talking about 100 milliseconds for linear algorithm vs. trillions of years for exponential algorithm.

GitHub: Fibonacci Code

Index

## JLTi Code Jam

At JLTi, I manage a monthly programming exercise. On the first week of every month, I set a programming problem and release it for all to solve by the end of the same month. We call it JLTi Code Jam, inspired by Google Code Jam.

We started it from Mar 2017 and so far we made it every month.

The programming problem is set in a way so that it can be solved using the data structures/algorithms discussed in the already conducted Friday Fun Sessions. The focus is on correctness, execution efficiency (time/space) and code quality.

Every JLTi Code Jam problem is published in this blog. The solution of a certain month’s JLTi Code Jam problem is discussed on the first Friday Fun Session on the following month.

JLTi Code Jam along with Friday Fun session is one of many endeavours as to how, we, mostly the engineers at JLTi, continuously learn, re-skill ourselves and sharpen our technical, programming and problem solving skills.

Finally, thank you all so much who participate in the JLTi Code Jam exercise, and encourage me to continue it. It is only you who made it a success so far.

Complete list of problems set so far.

## 18thFriday Fun Session – 19th May 2017

We use k-d tree, shortened form of k-dimensional tree, to store data efficiently so that range query, nearest neighbor search (NN) etc. can be done efficiently.

### What is k-dimensional data?

If we have a set of ages say, {20, 45, 36, 75, 87, 69, 18}, these are one dimensional data. Because each data in the array is a single value that represents age.

What if instead of only age we have to also store the salary for a person? The data would look like [{20, 1500}, {45, 5000}, {36, 4000}, {75, 2000}, {87, 0}, {18, 1000}]. This data is two dimensional as each data set contains two values. Similarly, if we add one more attribute to it, say education it would be a 3 dimensional data and so on.

### Why are we talking about efficiency?

Suppose, given a data point {43, 4650}, we want to know which person has a similar profile. In this particular example, it would be {45, 5000} whose age and salary both are close to this input. If we want the second closest person, it would be {36, 4000}. How did we find that? Well, we could iterate over the 6 data points and check against each of them. We would end up doing comparison against each of them. That is O(n) complexity. Not bad, but when we have millions of points it would be very expensive.

When we have just one dimension, instead of a linear search with O(n) complexity, we use Binary Search Tree (BST) with O(log2(n)) complexity. The difference is huge. For a million rows where linear search would take one million comparisons, binary search would take only 20 comparisons. This is because O(n) = O(1000, 000), meaning 1000, 000 comparisons and O(log2(n)) = O(log2(1000, 000)), meaning 20 comparisons. If each operation takes 1 millisecond BST would take 20 milliseconds, whereas linear search would take 1000 sec, almost 16 minutes. 20 milliseconds vs. 16 minutes.

### How do we split the points?

We can extend BST to do this. This is what Jon Louis Bentley created in 1975. K-d tree is called 2-d tree or k-d tree with 2-dimension when k = 2 and so on. In BST, at each level of the tree we split the data points based on the data value. Since, BST deals with just one dimension the question does not arise which dimension. But in k-d tree since we have more than one dimension. At each level we can choose to split the data based on only one dimension. So if we have 3 dimensions: x, y and z, at first level we split the data sets using x dimension. At 2nd level we do so using y dimension and at 3rd level we use z dimension. At 4th level we start again with x dimension and so on. Of course, we can continue splitting only if we have more data left. If we are splitting the points based on x dimension for a certain level then we call x the cutting dimension for this level.

### Where do the data points reside?

A k-d tree can have all the data points residing only in the leaf nodes. The intermediary nodes could be used to save the (non-data) splitting values. Alternatively, all nodes – internal and leaf, could save data points. In our case, we are saving data in all nodes.

### Balanced or Skewed

The above tree looks very symmetrical. That means, both the left sub-tree of right sub-tree having almost the same number of nodes. If the height of left and right sub-tree differs at max by 1 then it is called a balanced tree.

The more a tree is balanced the more efficient it is to do search and other operations on it. For example, if we have to do a search for a number in the above tree with height = 3, it would take at max 4 (height + 1) probes. If it were a skewed tree where most or all nodes reside on the same side, it would have taken 15 probes in the worst case, similar to a linear search.

### How can we build a balanced tree?

Let us start with an example set to walk through for the rest of the post. Say, we have 13 points in a two dimensional space. They are: (1, 3), (1, 8), (2, 2), (2, 10), (3, 6), (4, 1), (5, 4), (6, 8), (7, 4), (7, 7), (8, 2), (8, 5) and (9, 9) respectively.

Say, at level 1 the first dimension, say x is chosen as the cutting dimension. Since we want half the points to fall on the left side and the rest half on the right side we can simply sort (typically with O(n log2(n)) complexity) he data points on x dimension and chose the middle as the root. We make sure that we remain consistent in choosing for left side points whose cutting dimension value is less than the same for root and more than or equal to for the right.

In this example, if we sort the 13 points based on the x dimension values then the root would be (5, 4). So with (5, 4) being the root at level 1, the left side points would be: (1, 3), (1, 8), (2, 2), (2, 10), (3, 6), and (4, 1). And the right side points would be (6, 8), (7, 4), (7, 7), (8, 2), (8, 5) and (9, 9). We call the tree building procedure recursively for each half data sets. We also indicate that y dimension of the data point would be chosen as the cutting dimension for the next level sub-trees.

Now we have the following data points to build the left side tree with cutting dimension being y: (1, 3), (1, 8), (2, 2), (2, 10), (3, 6), and (4, 1). We can chose (3, 6) as the root, at level 2 after sorting them according to y dimension. The left side points would be (4, 1), (2, 2), and (1, 3) and the right side points would be (1, 8) and (2, 10).

At the end the tree would look like below:

### Bounding Box

We could visualize the points and the tree it in a different way. Let us put the 2 dimensional 13 points in x-y coordinate system.

The root (5, 4) at level 1 owns the whole bounding box. This root then divides the whole region into two bounding boxes: bounding box A and bounding box B, owned by second level roots (3, 6) and (8, 2) respectively.

Root (3, 6) would then divide the bounding box A into bounding box C and D owned by 3rd level roots (2, 2) and (2, 10) respectively. It does so using y as the cutting dimension meaning splitting the points inside A based on y dimension values.

Bounding box C rooted at (2, 2) is further divided into E and F, this time using x as the cutting dimension.

Bounding box E can be further divided into G and H using cutting dimension y but none of them has any point. Similarly, bounding box F can be further divided into I and J using cutting dimension y, once again none of them has any point.

Bounding box D can be divided into K and L using cutting dimension x. K having one point while L having no point inside it. K is further divided into two M and N using cutting dimension y having no points left for any of them.

Similarly bounding box B will be divided into smaller boxes.

The final bounding boxes are shown below. Even though bigger bounding boxes like A is not shown here, they are all present nonetheless. Only first level division of the B bounding box is shown where (7, 7) has split it into O and P.

### Nearest Neighbor Search

#### How many neighbors do we want?

We are interested to get k nearest neighbors, where k can be 1 2, 3 or any value. However, we will first see how to get the closest one point. It can then be easily extended to understand how to get more.

#### Points inside the same box of the query point are not necessarily the closest to query point

Suppose, we have to find the nearest neighbor of Query point Q = (4, 8) as shown below in red color. It falls inside bounding box D. But it is obvious that the closest point to Q does not fall within box D, rather it is inside Box B. Well, you can see point (6, 8) is the closest to Q. A person living near the Western border of Singapore is closer to a person living in the adjacent border of Malaysia than a person living in the eastern side of Singapore.

#### How to find the closest point?

We will extend the same binary search principle here. We start at root and then traverse down the tree finding the promising bounding boxes to search first and at the same time skip bounding boxes where the chance to get a closer point than the closest one to the query point found so far are thinner.

We will maintain the closest point (to Q) and minimum distance (distance between closest point and Q) found so far, at first they are null and infinite respectively. We start at root, with cutting dimension being x and do the following:

1. If we reach a null node return.
2. If the boundary box owned by the present root has no chance of having a point closer than minimum distance then return, meaning skip traversing that sub-tree altogether. We do so by checking the distance from Q to the bounding box. In two dimension case, it is Q to a rectangle (not a distance from Q to an actual point in the bounding box). This is how we prune search space.
3. If the present root is closer to Q than minimum distance, we save it as the closest point and also update the minimum distance.
4. Now we have two choices: traverse left sub-tree or traverse right sub-tree. We will compare the cutting dimension (at level 1 it is x, at level 2 it is y, at level 3 it is again x and so on) value of Q to that of root. If Q’s x is dimension value is smaller than that of root then we traverse left first, right second. So we are calling both of them but at a certain order with the hope that the first traversed sub-tree would give a closer point than any point the other side sub-tree could possibly offer. So next time when we would traverse the other sub-tree we can do a quick check and completely skip traversing that sub-tree. Something that might not materialize as well.

Let us walk through this particular example. At root (5, 4), closest point so far is null, minimum distance is also null. We set the root as the closet point and minimum distance (using commonly used Euclidean distance for continuous values) to ((x1 – x2)2 + (y1 –y2)2)1/2 = 4.12 (approx.). Now we have two bounding boxes A and B. The decision that we need to make is which one to traverse first? Q’s x dimension value 4 is smaller than root’s x dimension value that is 5. So we choose left first, right second. Both of them are to be called by using the cutting dimension y. The bounding box for each of the call is going to change. Well, we know each root owns a bounding box.

At the second call, root is (3, 6), bounding box is A. Distance from Q to A is zero as Q is within A. So we cannot skip traversing this sub-tree. Distance from (3, 6) to Q (4, 8) is 2.24, closer than existing minimum distance 4.12. Hence, we update our closest point to (3, 6) and minimum distance to 2.24. Next decision to make is again which side to traverse first. We have Q’s y value 8 that is more than present root’s y value 6. So we will traverse the right side first, left side second.

Next, the function is called with root, (2, 10), cutting dimension x and bounding box D. Distance between (2, 10) and Q (4, 8) = 2.83 that is larger than existing minimum distance. So we are not updating the closest point in this call. Next – choose which side to traverse first. Q’s x dimension 4 is bigger than root’s x dimension 2, hence right sub-tree is chosen first that is null anyway. So the call to it will return without doing anything.

Next call would be made with root (1, 8) that is 3 units away from Q. No improvement for the closest point. Also this root has no children. We have reached bottom of this side of the tree in our DFS search.

Next call is done with root (2, 2), far away from Q. But the bounding box owned by it is only 2 units away from Q. Hence, there is a chance that we might end up getting a closer point from this area. Hence, we cannot skip this tree. Right side to traverse first based on x dimension value comparison.

Root (4, 1), that is 8 units away from Q is called. It owns bounding box F that is 2 units away from Q. Once again cannot skip this area. Well, it has no child anyway.

Root (1, 3) is called that owns bounding box E that is 2.83 units away from Q, having no chance to offer any closer point. For the first time we can skip this area/sub-tree/bounding box.

We are done with the left side of level 1 root (5, 4). Now traverse right side.

Sub-tree with root (7, 7) owing bounding box O is called. Subsequently sub-tree rooted at (6, 8) would be called and that would be the closest point at distance 2.

#### How much search space did we prune?

We will skip traversing the left sub-tree rooted at (8, 2) that owns bounding box P. In terms of nodes we skipped only 4 nodes, 3 of them are rooted at (8, 2). Previously we skipped sub-tree rooted at (1, 3) as well. The green areas were pruned, meaning we did not search there. That was not quite efficient though!

#### How to get k nearest neighbors?

Instead of keeping a single closest point we could maintain a priority queue (max heap) to keep k (say 2, 3 or any number) closest points. The first k points would be en-queued anyway. Onwards, a new point, if better, would replace the worst of the closest point found so far. That way we can maintain the k nearest points easily.

### Too few points is a problem

If we have to construct a k-d tree with 20 dimensional data sets, we got to have around 220 data points. If we don’t have enough data then at many levels we will not have sufficient data to split. We will also end up with an unbalanced k-d tree where search and other operations would not be very efficient.

In general, we need k-d tree when we have higher dimensional data points. But when the dimension is too high other approaches might work better.

Index

## 17thFriday Fun Session – 12th May 2017

We use Bellman-Ford Algorithm to find the shortest path from a single source node/vertex (red color) to all destination nodes/vertices.

### Let’s use our intuition

Given that distance from city-1 (city-1 is node 1 here) to city-2 is 5, and from city-2 to city-3 is 6, if I have travel from city-1 to city-2 and city-3 respectively what would be the cheapest way to do so?

We can start from city-1 and reach city-2 at cost 5. Now that we have arrived at city-2 at cost 5, we can add cost 6 to it and reach city-3 via city-2. Thus, the shortest path from city-1 to city-2 and city-3 are respectively 5 and 11.

### Distance

Since city-1 is the source, we can add a self-loop on it with cost 0. That means, reaching city-1 from itself would cost 0. We also set that reaching city-2 and city-3 would cost infinity. We set so, because as of now we don’t know what would be the cost to reach there. So we put the maximum possible cost. Let’s call it distance. So we have distance [1] = 0, distance [2] = ∞ and distance [3] = ∞.

### Predecessor

Let’s also maintain another array, called predecessor to indicate the last node from which we arrived here. We set predecessor [1] = 0, predecessor [2] = 0, predecessor [3] = 0.  Since we have not arrived to city-2 and city-3 yet, we set the predecessor value for them to something invalid (0). For city-1, it is the source, can be set to 0 as well.

### Relaxation

Now we take each path/edge. We have two edges here. First one is from city-1 to city-2 with cost 5 and the second edge is from city-2 to city-3 with cost 3. Now let’s do what is called relaxation on each of the edges.

We see that using first edge we can arrive at city-2 from city-1 at a cost of 5 (distance [1] + cost of first edge). Since 5 is less than the existing distance of city-2 that is ∞, we update distance [2] to 5. We also note that, we arrived here from city-1 and hence got this new distance, that means we also update predecessor [2] = 1.

Now let’s do relaxation on edge 2. We see that distance [3] that is ∞ as of now can be improved by using edge 2. We set distance [3] = distance [2] + cost of edge 2 = 5 + 6 = 11. Since we arrived here from city-2, let’s update predecessor [3] = 2.

### Does order of edges for relaxation matter?

Now we are done with relaxation for all the edges once. First, we did the relaxation on first edge. Then we did the relaxation on the second edge. What would happen if the we changed the order? That means do the relaxation on the second edge first and then do it on the first edge.

Let’s do it. Start the relaxation afresh with new edge order. As of now we have predecessor [1] = 0, predecessor [2] = 0, predecessor [3] = 0. Also distance [1] = 0, distance [2] = ∞ and distance [3] = ∞.

We do relaxation on second edge (city-2 to city-3 at cost 6) first. We see that both distance [2] and distance [3] = ∞. Hence there is no chance to improve distance [3] since distance [2] + 6 = ∞ + 6 = ∞, that is no better than the existing distance [3] that also ∞. Hence relaxing second edge did not yield anything.

Let’s do relaxation on first edge. We know that would result in distance [2] = 5 and predecessor [2] = 1.

Well, at this point we see that we are done with relaxation on all the edges once and yet we have not found the shortest path to reach city-3.

So when shall we get the result?

### Iteration

That brings us to the next concept called iteration. Relaxation on all the edges once is called an iteration. So how many iterations do we need to get the shortest path to all destination nodes? Let’s use our intuition on the example that we are working on. We got 3 nodes. So if we have to reach from one end (say node 1) to the other end (say node 3), the maximum edges we might have to travel is 2, that is, the number of nodes minus 1. If we do the relaxation on the edges in an order that would choose the furthest edge from source (or the closest to the destination, in this case, second edge that is going from city-2 to city-3 at cost 6), we see that at each iteration we would increase the path (source node to furthest node) by at least one edge. And hence at 2 iterations we will certainly reach all reachable (I am saying reachable because all destinations might not be reachable) destinations.

Let’s continue our workout from where we left. Let’s start iteration 2 for the cases when we did the relaxation on second edge first. At this 2nd iteration, we again start with second edge. This time we can update distance [3] = distance [2] + 6 = 5 + 6 = 11. Predecessor [3] = 2.

We are done with2 iterations and we have found the result to reach from city-1 to both city-2 and city-3.

### Are all nodes reachable?

Let’s consider the below example, that is constructed by adding one more node 4.

We will see that distance [4] will remain ∞ and predecessor [4] will remain 0 (invalid) after |V| -1 = 4 – 1 = 3 iterations, where |V| is the number of nodes/vertices. This is because there is no incoming edge (path) to city-4. Hence city-4 is unreachable.

### Did order of edges for relaxation really matter?

We have seen that intermediate results (distance and predecessor values) might vary based on the order of edges we chose for relaxation but final result after all the iterations will still be the same. Hence the order of edges on which we do relaxation does not really matter (as far as final result is concerned).

### So after |V| – 1 iterations we have got the correct result?

Unfortunately not! Well, we did get the final result. But as of now we don’t know whether the result is valid or not. That sounds interesting. So, we have found the result and still we don’t know whether the result is correct/valid or not. So what is the issue? Well, let’s consider the below case.

I have added a third edge from city-3 to city-1. And the cost is -12. Negative cost? Why? We don’t answer the why question but let’s answer the what question. Cost -12 means reaching city-1 from city-4 would cost -1.

Let’s continue our workout where we have finished 2 iterations by considering the second edge first. Let’s assume we also considered the third edge and that third edge was considered at first for relaxation at each iteration. Distance [3] got a less than infinity value after 2nd iteration. Since we used 3rd edge at first, that means third edge was not used till 2nd iteration to update any distance for any node. That means the values (distance and predecessor) we got last time would be the same value even with the presence of third edge after 2 iterations.

Since we just added an extra edge (3rd edge) but no new nodes, the number of iterations we have to do still remains 2. That also means the result we found so far still valid in this case. But is the result (distance [2] = 5 and distance [3] = 11) correct with this new situation?

### Negative cycle

Now that you arrived at city-3 at cost 11, you can go to city-1 at cost 11 + (-12) = -1 and then city-2 at -4 and so on. The more you travel, the less cost you would incur. And hence the shortest path found after 2 iterations are not valid.

So how do we find that the result is invalid? Well after we are done with |V|-1 iterations, we have to do one more iteration that is the |V|th one (a cycle involving |V| nodes can be found with |V| edges). If that changes distance value for any node that means there exists a negative cycle (a cycle whose edges (costs) sum to a negative value). When there is negative cycle present in a graph then the answer found is invalid.

### Negative edge vs. negative cycle

Does negative edge means negative cycle? Does presence of negative edge mean no answer can be found?

Negative edge is fine with Bellman-Ford as in the above example. A correct solution can still be found. A correct solution cannot be found when there is a negative cycle. But Bellman-Ford can detect a negative cycle and in that case it can indicate that a correct solution is not found.

### Are all iterations required?

Not really. When we did the relaxation on the first edge first, we already found the shortest paths to both city-2 and city-3. How do we know? Well, at iteration 2 we would have found that no distance got updated. If an iteration does not change any distance value then we can terminate the algorithm there and return valid result. Because in that case, subsequent iterations are not going to change anything. It also means there is no negative cycle.

### The shortest path sequence

We can use the predecessor array recursively to get the shortest path sequence. For example, earlier after 2 iterations we got the following result.

distance [1] = 0, distance [2] = 5, distance [3] = 11

predecessor [1] = 0, predecessor [2] = 1, predecessor [3] = 2

If we want to find the shortest path sequence to city-3, we can find the predecessor [3] that is 2, recursively we can check the predecessor [2] that is 1 and that equals to source node, that is city-1. So we stop and the sequence is city-1 to city-2 to city-3.

### Distance for a particular node can be updated more than once in an iteration

In this example above we have two nodes. We have to do one iteration. We have two edges: first with cost 5, second with cost 2. Source is node 1. If we relax the first edge then distance [2] will be 5. Subsequent relaxation on second edge would result the node 2 distance to be updated again with 2 (because, existing distance 5 > (0 + 2)). We see that node 2 distance got updated twice within the same iteration.

### The algorithm

Now that we have done with the workout, let’s write down the algorithm.

```Function BellmanFord()
{
input = G {V, E};
distance[] = ∞;
predecessor[] = -1;
distance[sourceNode] = 0;

for i = 1 to |V|-1
{
valueChanged = false;
for j = 1 to |E|
valueChanged = Relax (E[j]) || valueChanged;

if(!valueChanged)
return Result();
}

for j = 1 to |E|
if(Relax(E[j])
print ‘negative cycle detected, solution not possible’;
}

Function Relax (e)
{
if(distance[e.to]) > distance[e.from] + e.cost)
{
distance[e.to] = distance[e.from] + e.cost;
predecessor[e.to] = e.from;
return true;
}

return false;
}

Function Result()
{
print ‘success’;
print distance[];
print predecessor[];
}```

### The complexity

For each iteration (number of vertices – 1), we are iterating over all the edges. That means the complexity is O(|V|. |E|).

GitHub: Manipulating Money Exchange

Index

## 3rdJLTi Code Jam – May 2017

Input: price = [961, 984.5, 965, 988.5, 956.5]

Explanation: Now that your bank accounts are flooding with April bonus money and you are contemplating to invest in stock market why not join me in doing some analysis first? After all, we are engineers flooded with data (quite a lot of them are free) and the capability to analyse them. I opened JLT’s historical stock price from yahoo finance. It’s amazing! As part of the analysis, you know, we have to do a tremendous amount of work. As a start, I wanted to focus on when to buy and when to sell a stock so that I can make the most profit. For example, I took the data from 19th Dec 2016 to 23rd Dec 2016 as specified above. I did some manual calculation and found that had I bought on 19th Dec 2016 at £961 and sold on 22nd Dec 2016 £988.5, I could make the most profit, £27.5.

I also checked if I had to make the most profit by buying and selling a single JLT share once within 2016 , I could do so by buying on 9th Feb 2016 at £776.5 and selling on 11th Apr 2016 at £1070, making a whopping £293.5 profit! I have not put the data here as one year’s data is too huge to fit here. You can collect and verify it by downloading it as an excel file from the above yahoo link.

Output: Buy on day 1 at £961 and sell on day 4 at £988.5, making £27.5

I also checked the values from 15th Jan 2016 to 21st Jan 2016 (excluding 16th Jan 2016 and 17th Jan 2016 when stock market was closed) and the price looked like the below.

Input: price = [890, 890, 853.5, 828.5, 809]

You can see it kept on dropping and there was no way to make money in this period.

Task: As you realize, as part of the bigger data analysis work that we need to do, it is a small part. We got so much data for hundreds of companies. Hence, it is essential that we do it efficiently. To be precise, I am looking for a solution more time efficient than O(n2).

Index

## 1stJLTi Code Jam – Mar 2017

Input: 1, 1, 2, 2, 2, 567, 567, 10000076, 4, 2, 3, 3

Explanation: There are 12 people listed above. They belong to 6 teams (Team 1, Team 2, Team 3, Team 567, Team 10000076, and Team 4). As you can see people are identified in the list by the team number.

Output: 1, 2, 1, 2, 4, 2, 567, 3, 567, 100076, 2, 3

As you can see, the output has rearranged the team members in way that no two members from the same team standing next to each other.

Input: 1, 1, 1, 1, 2

Output: It is not possible to rearrange them.

Task: You have to write a program that can rearrange even billions of such team members belonging to millions of teams very fast. If the input is such that it is not possible to rearrange then the output should be: It is not possible to rearrange them. A correct solution is not sufficient. The algorithm has to be efficient, otherwise the output for big data 🙂 will not come.

Index

## 2ndJLTi Code Jam – Apr 2017

Input: Capacity = 125, w = [45, 25, 80, 100, 125]

Explanation: This year, RC has taken all JLT Asia employees to Noland for the company trip. As the name implies there is not much land in Noland, it is river everywhere. When we have to cross such a river having only one boat with a certain capacity (in the above example 125 Kg), Warren Downey, our Deputy CEO approaches RC and asks us to quickly divide the people so that each trip of the boat carries people exactly to its maximum capacity, 125 Kg in this example. He shows the example above and works out the below output that he desires.

Output: {45, 80}, {25, 100}, {120}

When RC team pointed out what would happen for a scenario when we have a case like Capacity = 120, w = [40, 20, 80, 100, 120, 70]. Warren informs us we always utilize our resources to its maximum capacity. No compromise. We will not cross the river and will change the tour itinerary.

Output: No crossing, change itinerary.

Task: When I woke up an hour ago from my afternoon nap with a lot of stress, I realized that the tour was just a bad dream. I started feeling relaxed. But the problem got into my head and now it is itching everywhere inside it. In this situation, I realize, I can spread the itching to my JLTi friends in Singapore and Mumbai as well.

You can imagine there is a boatman and his weight is out of consideration. The input capacity is only applicable for the passengers. The input w array is holding only the passenger weights. In short, you can ignore the boatman.

GitHub: Company Tour to Noland

Index