JOIN
 Match Editorial
SRM 272
Saturday, November 19, 2005

Match summary

With twelve coders in Div1 solving all three problems, this match was easier than some, but not without its difficulties. radeye was able to come up with two successful challenges to pass Eryx and take 3rd in the match, but this wasn't enough to surpass tomek who got a successful challenge of his own to keep him in 2nd. But Petr, who had the fastest correct submission for both the 500 and 1000, easily took first.

In Div2, newcomers Nevo and simonenko took first and second place, despite being in the same room. In fact simonenko attempted a challenge on Nevo's 1000 point problem, but to no avail as Nevo took the room and division win.

# The Problems

HammingDistance
Used as: Division Two - Level One:
 Value 250 Submission Rate 415 / 445 (93.26%) Success Rate 364 / 415 (87.71%) High Score lllllll for 249.74 points (0 mins 55 secs) Average Score 214.07 (for 362 correct submissions)

Given two strings of equal length, it is quite simple to calculate their Hamming distance. All you have to do is loop through each character of the first string and compare it to the analogous character in the second string, if they don't match then increment a counter. The runtime of this process is proportional to the length of the strings.

To find the minimum of Hamming distances of all pairs of strings you can calculate the Hamming distance between all pairs of strings. If there are N strings each with length L, then there are N(N-1)/2 pairs (be sure not to compute a string's distance to itself), so the total runtime will be O(N2 L). Since N and L are both bounded by 50, this is around 125,000 operations, which is small enough for this brute force method to work.

FewestFactors
Used as: Division Two - Level Two:
 Value 500 Submission Rate 286 / 445 (64.27%) Success Rate 176 / 286 (61.54%) High Score Xepo for 483.16 points (5 mins 20 secs) Average Score 325.78 (for 176 correct submissions)
Used as: Division One - Level One:
 Value 250 Submission Rate 363 / 374 (97.06%) Success Rate 301 / 363 (82.92%) High Score Zuza for 248.59 points (2 mins 8 secs) Average Score 214.36 (for 301 correct submissions)

This is the first problem where permutations come in handy. We know that each of the elements in digits must be used exactly once, so the only decision left up to us is the order the digits are put in. Every ordering of the digits is a permutation, thus the problem is to determine the permutation of digits which yields the fewest number of factors. C++ coders could use the next_permutation function, other languages had to generate the permutations using some other method (it can be done recursively with a depth first search, see the Round Tables for details).

Once you have a permutation of digits, you must determine how many factors it has. First transform the permutation into a number by summing each digit times the proper power of ten. Once you have this candidate number, you can loop through each positive integer less than the candidate to see if it is a factor. A number is a factor if when you divide the candidate by it you get a remainder of 0. If the candidate has the fewest factors seen so far, keep it.

With at most 5 digits there can be up to 5! = 120 permutations. With each permutation you have to check at most 99,999 possible factors (since each number has at most 5 digits). This means that the entire process takes on the order of 120*100,000 = 12,000,000 operations. This is fast enough to pass.

Another way to count the factors is to note that for every factor less than the square root of the candidate there is exactly one factor greater than the square root. Thus you really only need to loop up to the square root counting each factor twice (and if the candidate is a square, only count the square root once). This cuts down significantly on the runtime.

VectorPolygon
Used as: Division Two - Level Three:
 Value 1000 Submission Rate 60 / 445 (13.48%) Success Rate 25 / 60 (41.67%) High Score Nevo for 761.27 points (17 mins 4 secs) Average Score 526.40 (for 25 correct submissions)

One of the properties of a Convex polygon is that the angle each edge makes with a horizontal line increases as you go around it (and decreases as you go around it the other way). In fact, this is the property that the Graham scan algorithm for finding a convex hull is based on. Knowing this, you can sort all of the vectors by the angle they make with the horizontal. Then try each subset of vectors to see if it makes a polygon. The images included with the problem should give one the intuition that the vectors will only make a polygon if the sum of the vectors is 0. To find the area of the polygon you should check out the geometry article in the Educational Content section. Thus for each subset, if the sum of the vectors is 0 and the area the polygon makes is larger than any that you've seen then store that area as your best.

A way of doing this problem without using the above property makes use of permutations. In this method you try every possible permutation of the vectors. Then start with the first vector in the permutation and keep on adding vectors until the sum is 0. When this happens you know you have a polygon, so find its area. The only drawback of this method is that you aren't sure that the polygon is convex. Luckily this doesn't matter, because if you do find a polygon then some reordering of its sides is guaranteed to be convex (namely when the sides are sorted by angle with the horizontal). Further when the sides make a convex polygon that area will be larger (intuitively the concave parts of the polygon take area away from what could be).

RoundTable
Used as: Division One - Level Two:
 Value 500 Submission Rate 160 / 374 (42.78%) Success Rate 83 / 160 (51.88%) High Score Petr for 467.77 points (7 mins 33 secs) Average Score 265.95 (for 83 correct submissions)

This problem, like many other combinatorial problems, can be solved with Dynamic Programming. The difficult part is coming up with the right formulation of what the sub-problem is, a way to describe the state space. To find what the state space should be more analysis of the problem is necessary.

The problem would be quite a bit easier if the table wasn't round. While we can't flatten the table we can make an observation which is just as good. Think of some distinguished member of company A (the CEO for instance). The CEO of company A has to sit in some chair, so lets label that chair 0 and then label the chair to it's right 1, the next one 2, and so on. Now the problem is how to seat the remaining people in the linear sequence of chairs1, 2... chairs-1. The only restrictions are you can't put a member of company B into the first or last minDistance-1 chairs, because the CEO of A is in chair 0. Also, once we get the result we must multiply it by chairs since any rotation of this solution is a distinct valid solution.

What else is part of our state space? We need to know which chair we want to try to place the next person at, call it currentChair. Since we can only place a member of company A if we know she is far enough away from the closest current member of company B, we also have to keep track of the previousBrep (which can have a special value indicating no B has been placed yet). Likewise we need to keep track of previousArep. Finally we need to know how many representatives of company A remain, remainingAreps, and similarly remainingBreps.

Thus the entire state space is described by the quintuple (currentChair,previousArep,previousBrep,remainingAreps,remainingBreps). Since the first three variables can take values from 0 to 50 and the last two can take values from 0 to 5, there are a total of 513 * 62 states, which is about 5 million. If only a constant amount of work is done for the recurrence, then this solution will be fast enough.

In each state, there are only three things that can be done: 1) place a member of company A into currentChair, 2) place a member of company B into currentChair, or 3) leave currentChair empty. In the first case, you must make sure that a representative from company A can indeed be placed. This means that currentChair must be at least minDistance away from previousBrep and there must be at least 1 remainingAreps. If these conditions are met, then add to the number of solutions of the current state the number of solutions to the state where you place a representative of A into CurrentChair times the number of representatives of A that remain, since you can put any of them into the seat. The second case is similar to the first case, but check if you can place someone from company B. Also remember you can't place a member of B into the last minDistance-1 chairs. The last case is the simplest because no checks need to be made.

The base case of the dynamic programming is when you get to the last chair. If all of the representatives have been placed then this is a unique valid solution so its value is 1, otherwise it is not a solution so its value is 0. So each state can be evaluated in constant time and this method will work.

ManhattanDistance
Used as: Division One - Level Three:
 Value 1000 Submission Rate 40 / 374 (10.70%) Success Rate 19 / 40 (47.50%) High Score Petr for 761.32 points (17 mins 4 secs) Average Score 573.66 (for 19 correct submissions)

One of the first lessons in Euclidean geometry is that the shortest path between any two points is a straight line. If that were true in Manhattan then this would be a truely simple problem, but unfortunately buildings get in the way. So the only reason you would avoid a straight line path is if the path intersects some building. In such a case you have to "go around" the building, but you want to stay as close to the straight line path as possible. This means that you'll actually travel to one of the corners of the building (and you'll want to take a straight line path to the corner). So any path from the start to the target will involve a sequence of linear segments with the endpoints on either the starting point, the target point, or a corner to one of the buildings. This limits the amount of possible paths substantially.

With only 26 avenues and 50 streets, there are at most 1,300 buildings, so 5,200 corners. So a graph can be constructed with 5,202 nodes with an edge between two nodes if the straight line between the two points does not intersect a building. Once this graph is constructed a shortest path algorithm, such as Dijsktra's, can be used to find the length of the shortest path.

The construction of this graph is made even easier if you note that the corner of each building can only "see" the corners of buildings along the same avenue or street. This is because of that strange constraint that width will be no greater than half of distance.

One thing to notice which could improve your implementation is that from the regularity of the buildings in the city, one should always make each move to a corner which is closer to the target in straight line distance. Using this you can use dynamic programming where the state is which corner you are at and you populate the table starting with the closest points to the start.

By Ryan
TopCoder Member