Let's say we have some array of boolean values:
A = [0 1 1 1 1 0 0 0 0 0 0 0 1 1 1 0 0 0 1 1 0 1 1 1 1 0 0 0 ... 0]
The array is constructed by performing classification on a stream of data. Each element in the array corresponds to the output of a classification algorithm given a small "chunk" of the data. An answer may include restructuring the array to make parsing more efficient.
The array is pseudo random in the sense that groups of 1's and 0's tend to exist in bunches (but not necessarily always).
Given some index, i, what is the most efficient way to find the group of at least n zeros closest to A[i]? For the easy case, take n = 1.
EDIT: Groups should have AT LEAST n zeros. Again, for the easy case, that means at least 1 zero.
EDIT2: This search will be performed o(n) times, where n is size of the array. (Specifically, its n/c, where c is some fixed duration.
In this solution I organize the data so that you can use a binary search O(log n) to find the nearest group of at least a certain size.
I first create groups of zeros from the array, then I put each group of zeros into lists containing all groups of size s or larger , so that when you want to find the nearest group of s s or more then you just run a binary search in the list that has all groups with a size of s or greater.
The downside is in the pre-processing of putting the groups into the lists, with O(n * m) (I think, please check me) time and space efficiency where n is the number of groups of zeros, and m is the max size of the groups, though in reality the efficiency is probably better.
Here is the code:
public static class Group {
final public int x1;
final public int x2;
final public int size;
public Group(int x1, int x2) {
assert x1 <= x2;
this.x1 = x1;
this.x2 = x2;
this.size = x2 - x1 + 1;
}
public static final List<Group> getGroupsOfZeros(byte[] arr) {
List<Group> listOfGroups = new ArrayList<>();
for (int i = 0; i < arr.length; i++) {
if (arr[i] == 0) {
int x1 = i;
for (++i; i < arr.length; i++)
if (arr[i] != 0)
break;
int x2 = i - 1;
listOfGroups.add(new Group(x1, x2));
}
}
return Collections.unmodifiableList(listOfGroups);
}
public static final Group binarySearchNearest(int i, List<Group> list) {
{ // edge cases
Group firstGroup = list.get(0);
if (i <= firstGroup.x2)
return firstGroup;
Group lastGroup = list.get(list.size() - 1);
if (i >= lastGroup.x1)
return lastGroup;
}
int lo = 0;
int hi = list.size() - 1;
while (lo <= hi) {
int mid = (hi + lo) / 2;
Group currGroup = list.get(mid);
if (i < currGroup.x1) {
hi = mid - 1;
} else if (i > currGroup.x2) {
lo = mid + 1;
} else {
// x1 <= i <= x2
return currGroup;
}
}
// intentionally swapped because: lo == hi + 1
Group lowGroup = list.get(hi);
Group highGroup = list.get(lo);
return (i - lowGroup.x2) < (highGroup.x1 - i) ? lowGroup : highGroup;
}
}
NOTE: GroupsBySize can be improved, as described by #maraca to only contain a list of Groups per each distinct group size. I'll update tomorrow.
public static class GroupsBySize {
private List<List<Group>> listOfGroupsBySize = new ArrayList<>();
public GroupsBySize(List<Group> groups) {
for (Group group : groups) {
// ensure internal array can groups up to this size
while (listOfGroupsBySize.size() < group.size) {
listOfGroupsBySize.add(new ArrayList<Group>());
}
// add group to all lists up to its size
for (int i = 0; i < group.size; i++) {
listOfGroupsBySize.get(i).add(group);
}
}
}
public final Group getNearestGroupOfAtLeastSize(int index, int atLeastSize) {
if (atLeastSize < 1)
throw new IllegalArgumentException("group size must be greater than 0");
List<Group> groupsOfAtLeastSize = listOfGroupsBySize.get(atLeastSize - 1);
return Group.binarySearchNearest(index, groupsOfAtLeastSize);
}
}
public static void main(String[] args) {
byte[] byteArray = null;
List<Group> groups = Group.getGroupsOfZeros(byteArray);
GroupsBySize groupsBySize = new GroupsBySize(groups);
int index = 12;
int atLeastSize = 5;
Group g = groupsBySize.getNearestGroupOfAtLeastSize(index, atLeastSize);
System.out.println("nearest group is (" + g.x1 + ":" + g.x2 + ") of size " + g.size);
}
If you have n queries on an array of size n, then the naive approach would take O(n^2) time.
You can optimize this by incorporating the observation that the number of distinct group sizes is in the order of sqrt(n), because the most distinct group sizes we get if we have one group of size 1, one of size 2, one of size 3 and so on, we know that 1 + 2 + 3 + ... + n is n * (n + 1) / 2, so in the order of n^2, but the array has size n, so the number of distinct group sizes is in the order of sqrt(n).
create an integer array of size n to denote which group sizes are present how many times
create a list for the 0-groups, each element should contain the group size and starting index
scan the array, add the 0-groups to the list and update the present group sizes
create an array for the different group sizes, each entry should contain the group size and an array with the start indices of the groups
create an integer array or a map which tells you which group size is at which index by scanning the array of the present group sizes
go through the list of 0-groups and fill the start index arrays created at 4.
We end up with an array which takes O(n) space, takes O(n) time to create and contains all present group sizes in order, additionally each entry has an array with the start indices of the groups of that size.
To answer a query we can do a binary search on the start indices of all groups greater or equal than the given minimum group size. This takes O(log(n)*sqrt(n)) and we do it n times, so over all it would take O(n*log(n)*sqrt(n)) = O(n^1.5*log(n)) which is better than O(n^2).
I think you can get it down to O(n^1.5) by creating a structure which has all distinct group sizes but contains not only the groups of that size, but also the groups that are bigger than that size. This would be the time complexity to create the structure and answering all the n queries would be faster O(n*log(sqrt(n))*log(n)) I think, so it doesn't matter.
example:
[0 1 1 1 1 0 0 0 0 0 0 0 1 1 1 0 0, 1, 0, 0] -- 0 indexed array
hashmap = {1:[0], 2:[15, 18], 7:[5]}
search(i = 7, n = 2) {
binary search in {2:[15, 18], 7:[5]}
return min(15, 5)
}
what is the most efficient way to find the group of at least n zeros closest to A[i]
If we are not limited in preprocessing time and resources, the most efficient way would seem to be O(1) time and O(n * sqrt n) space, storing the answers to all possible queries. (To accomplish that, run the algorithm below with a list of all possible queries, that is each distinct zero-group size in the array paired with each index.)
If we are provided with all the n / c queries at once, we can produce the complete result set in O(n log n) time.
Traverse once from the left and once from the right. For each traversal, start with a balanced binary tree with our queries, sorted by zero-group-size (the n in the query), where each node has a sorted list of the query indexes (all is with this particular n).
At each iteration, when a zero-group is registered, update all queries with n equal and lower than this zero-group size, removing all equal and lower indexes from the node and keeping the records for them (since the index list is sorted, we just remove the head of the list while it's equal or lower than the current index), and storing the current index of the zero-group in the node as well (the "last seen" zero-group-index). If no is are left in the node, remove it.
After the traversal, assign each node's "last seen" zero-group-index to any remaining is in that node. Now we have all the answers for this traversal. (Any queries left in the tree have no answer.) In the opposite traversal, if any query comes up with a better (closer) answer, update it in the final record.
It's a Google interview question. There's a list of "T" and "F" only. All denotes a position such that T means position is occupied by a flower pot and F means pot is not there, so you can put another pot at this position. Find the number of pots that can be placed in a given arrangement such that no two pots are adjacent to each other(they can be adjacent in the given arrangement). If a position at the beginning is unoccupied then a pot can be placed if second position is also unoccupied and if the last position is unoccupied than a pot can be placed if second last position is also unoccupied. For ex.
TFFFTFFTFFFFT - returns 2
FFTTFFFFFTTFF - returns 4
I tried solving it by looking at adjacent values for every position with value F. Increased the counter if both adjacent positions were F and set this position as T. I need a better solution or any other solution(if any).
Let's analyse what has to be done.
So first we probably need to visit and examine each place. That suggests loop of some sort. E.g.:
for (int i = 0; i < myPlaces.Length; ++i)
When we are at a spot we have to check if it's occupied
if (place[i] == 'F')
but that's not enough to place the flower pot there. We have to check if the next and previous place is free
place[i-1]
place[i+1]
If all tree contain F you can put the flower pot there and move to next field
Now, we also have some exceptions from the rule. Beginning and end of the list. So you have to deal with them separately. E.g
if (i == 0)
{
// only check current position and next position
}
if (i == myPlaces.Length - 1) // minus 1 because indexing usually starts from 0
{
// only check current position and previous position
}
After that you can perform the checks mentioned previously.
Now let's think of the input data. Generally, it's a good habit not to modify the input data but make a copy and work on the copy. Also some data structures work better than the others for different tasks. Here you can use simple string to keep entry values. But I would say an array of chars would be a better option because then, when you find a place where you can put a flower pot you can actually replace the F with the T in an array. Then when you move to new spot your data structers knows that there is already a pot in the previous position so your algorithm won't put an adjacent one.
You would not be able to do that with string as strings are immutable and you would need to generate a new string each time.
Note that it's only a naive algorithm with a lot of scope for improvement and optimization. But my goal was rather to give some idea how to approach this kind of problems in general. I'll leave implementing of the details to you as an afternoon exercise before targeting a job at Google.
You may be able to do this with a modified Mergesort. Consider the flowerpots that can be placed in the singletons, then the flowerpots that can be placed in the doubleton merges of those singletons, up the tree to the full arrangement. It would complete in O(n lg n) for a list of n flowerpots.
There is certainly a way to do this with a modified Rod Cutting algorithm with complexity O(n^2). The subproblem is whether or not an open "false set" exists in the substring being considered. The "closed false sets" already have some maximum value computed for them. So, when a new character is added, it either increases the amount of flowerpots that can be inserted, or "locks in" the maximum quantity of available flowerpots for the substring.
Also, you know that the maximum flowerpots that can be placed in a set of n open positions bound by closed positions is n - 2 (else n-1 if only bracketed on one side, i.e. the string begins or ends with a "false set". The base condition (the first position is open, or the first position is closed) can calculated upon reaching the second flowerpot.
So, we can build up to the total number of flowerpots that can be inserted into the whole arrangement in terms of the maximum number of flowerpots that can be inserted into smaller subarrangements that have been previously calculated. By storing our previous calculations in an array, we reduce the amount of time necessary to calculate the maximum for the next subarrangement to a single array lookup and some constant-time calculations. This is the essence of dynamic programming!
EDIT: I updated the answer to provide a description of the Dynamic Programming approach. Please consider working through the interactive textbook I mentioned in the comments! http://interactivepython.org/runestone/static/pythonds/index.html
I would approach the problem like this. You need FFF to have one more pot, FFFFF for two pots, etc. To handle the end cases, add an F at each end.
Because this is very similar to a 16-bit integer, the algorithm should use tricks like binary arithmetic operations.
Here is an implementation in Python that uses bit masking (value & 1), bit shifting (value >>= 1) and math ((zeros - 1) / 2) to count empty slots and calculate how many flower pots could fit.
#value = 0b1000100100001
value = 0b0011000001100
width = 13
print bin(value)
pots = 0 # number of flower pots possible
zeros = 1 # number of zero bits in a row, start with one leading zero
for i in range(width):
if value & 1: # bit is one, count the number of zeros
if zeros > 0:
pots += (zeros - 1) / 2
zeros = 0
else: # bit is zero, increment the number found
zeros += 1
value >>= 1 # shift the bits to the right
zeros += 1 # add one trailing zero
pots += (zeros - 1) / 2
print pots, "flower pots"
The solution is really simple, check the previous and current value of the position and mark the position as plantable (or puttable) and increment the count. Read the next value, if it is already is planted, (backtrack and) change the previous value and decrement the count. The complexity is O(n). What we really want to check is the occurrence of 1001. Following is the implementation of the algorithm in Java.
public boolean canPlaceFlowers(List<Boolean> flowerbed, int numberToPlace) {
Boolean previous = false;
boolean puttable = false;
boolean prevChanged = false;
int planted = 0;
for (Boolean current : flowerbed) {
if (previous == false && current == false) {
puttable = true;
}
if (prevChanged == true && current == true) {
planted--;
}
if (puttable) {
previous = true;
prevChanged = true;
planted++;
puttable = false;
} else {
previous = current;
prevChanged = false;
}
}
if (planted >= numberToPlace) {
return true;
}
return false;
}
private static void canPlaceOneFlower(List<Boolean> flowerbed, FlowerBed fb) {
boolean result;
result = fb.canPlaceFlowers(flowerbed, 1);
System.out.println("Can place 1 flower");
if (result) {
System.out.println("-->Yes");
} else {
System.out.println("-->No");
}
}
private static void canPlaceTwoFlowers(List<Boolean> flowerbed, FlowerBed fb) {
boolean result;
result = fb.canPlaceFlowers(flowerbed, 2);
System.out.println("Can place 2 flowers");
if (result) {
System.out.println("-->Yes");
} else {
System.out.println("-->No");
}
}
private static void canPlaceThreeFlowers(List<Boolean> flowerbed, FlowerBed fb) {
boolean result;
result = fb.canPlaceFlowers(flowerbed, 3);
System.out.println("Can place 3 flowers");
if (result) {
System.out.println("-->Yes");
} else {
System.out.println("-->No");
}
}
private static void canPlaceFourFlowers(List<Boolean> flowerbed, FlowerBed fb) {
boolean result;
result = fb.canPlaceFlowers(flowerbed, 4);
System.out.println("Can place 4 flowers");
if (result) {
System.out.println("-->Yes");
} else {
System.out.println("-->No");
}
}
public static void main(String[] args) {
List<Boolean> flowerbed = makeBed(new int[] { 0, 0, 0, 0, 0, 0, 0 });
FlowerBed fb = new FlowerBed();
canPlaceFourFlowers(flowerbed, fb);
canPlaceThreeFlowers(flowerbed, fb);
flowerbed = makeBed(new int[] { 0, 0, 0, 1, 0, 0, 0 });
canPlaceFourFlowers(flowerbed, fb);
canPlaceThreeFlowers(flowerbed, fb);
canPlaceTwoFlowers(flowerbed, fb);
flowerbed = makeBed(new int[] { 1, 0, 0, 1, 0, 0, 0, 1 });
canPlaceFourFlowers(flowerbed, fb);
canPlaceThreeFlowers(flowerbed, fb);
canPlaceTwoFlowers(flowerbed, fb);
canPlaceOneFlower(flowerbed, fb);
}
My solution using dynamic programming.
ar is array in the form of ['F','T','F'].
import numpy as np
def pot(ar):
s = len(ar)
rt = np.zeros((s,s))
for k in range(0,s):
for i in range(s-k):
for j in range(i,i+k+1):
left = 0
right = 0
if ar[j] != 'F':
continue
if j-1 >= i and ar[j-1] == 'T':
continue
else:
left = 0
if j+1 <= i+k and ar[j+1] == 'T':
continue
else:
right = 0
if j-2 >= i:
left = rt[i][j-2]
if j+2 <= i+k:
right = rt[j+2][i+k]
rt[i][i+k] = max(rt[i][i+k], left+right+1)
return rt[0][len(ar)-1]
My solution written in C#
private static int CheckAvailableSlots(string str)
{
int counter = 0;
char[] chrs = str.ToCharArray();
if (chrs.FirstOrDefault().Equals('F'))
if (chrs.Length == 1)
counter++;
else if (chrs.Skip(1).FirstOrDefault().Equals('F'))
counter++;
if (chrs.LastOrDefault().Equals('F') && chrs.Reverse().Skip(1).FirstOrDefault().Equals('F'))
counter++;
for (int i = 1; i < chrs.Length - 2; i++)
{
if (chrs[i - 1].Equals('T'))
continue;
else if (chrs[i].Equals('F') && chrs[i + 1].Equals('F'))
{
chrs[i] = 'T';
counter++;
i++;
}
else
i++;
}
return counter;
}
// 1='T'
// 0='F'
int[] flowerbed = new int[] {1,0,0,0,0,1};
public boolean canPlaceFlowers(int[] flowerbed, int n) {
int tg = 0;
for (int i = 0, g = 1; i < flowerbed.length && tg < n; i++) {
g += flowerbed[i] == 0 ? flowerbed.length - 1 == i ? 2 : 1 : 0;
if (flowerbed[i] == 1 || i == flowerbed.length - 1) {
tg += g / 2 - (g % 2 == 0 ? 1 : 0);
g = 0;
}
}
return tg >= n;
}
Most of these answers (unless they alter the array or traverse and a copy) dont consider the situation where the first 3 (or last 3) pots are empty. These solutions will incorrectly determine that FFFT will contain 2 spaces, rather than just one. We therefore need to start at the third element (rather than then second) and end at index length - 3 (rather than length - 2). Also, while looping through the array, if an eligible index is found, the index just be incremented by 2, otherwise TTFFFFT would give 2 available plots instead of one. This is true unless you alter the array while looping or use a copy of the array and alter it.
Edit: this holds true unless the question is how many spaces are available for planting, rather than how many total plants can be added
I was recently given this interview question and I'm curious what a good solution to it would be.
Say I'm given a 2d array where all the
numbers in the array are in increasing
order from left to right and top to
bottom.
What is the best way to search and
determine if a target number is in the
array?
Now, my first inclination is to utilize a binary search since my data is sorted. I can determine if a number is in a single row in O(log N) time. However, it is the 2 directions that throw me off.
Another solution I thought may work is to start somewhere in the middle. If the middle value is less than my target, then I can be sure it is in the left square portion of the matrix from the middle. I then move diagonally and check again, reducing the size of the square that the target could potentially be in until I have honed in on the target number.
Does anyone have any good ideas on solving this problem?
Example array:
Sorted left to right, top to bottom.
1 2 4 5 6
2 3 5 7 8
4 6 8 9 10
5 8 9 10 11
Here's a simple approach:
Start at the bottom-left corner.
If the target is less than that value, it must be above us, so move up one.
Otherwise we know that the target can't be in that column, so move right one.
Goto 2.
For an NxM array, this runs in O(N+M). I think it would be difficult to do better. :)
Edit: Lots of good discussion. I was talking about the general case above; clearly, if N or M are small, you could use a binary search approach to do this in something approaching logarithmic time.
Here are some details, for those who are curious:
History
This simple algorithm is called a Saddleback Search. It's been around for a while, and it is optimal when N == M. Some references:
David Gries, The Science of Programming. Springer-Verlag, 1989.
Edsgar Dijkstra, The Saddleback Search. Note EWD-934, 1985.
However, when N < M, intuition suggests that binary search should be able to do better than O(N+M): For example, when N == 1, a pure binary search will run in logarithmic rather than linear time.
Worst-case bound
Richard Bird examined this intuition that binary search could improve the Saddleback algorithm in a 2006 paper:
Richard S. Bird, Improving Saddleback Search: A Lesson in Algorithm Design, in Mathematics of Program Construction, pp. 82--89, volume 4014, 2006.
Using a rather unusual conversational technique, Bird shows us that for N <= M, this problem has a lower bound of Ω(N * log(M/N)). This bound make sense, as it gives us linear performance when N == M and logarithmic performance when N == 1.
Algorithms for rectangular arrays
One approach that uses a row-by-row binary search looks like this:
Start with a rectangular array where N < M. Let's say N is rows and M is columns.
Do a binary search on the middle row for value. If we find it, we're done.
Otherwise we've found an adjacent pair of numbers s and g, where s < value < g.
The rectangle of numbers above and to the left of s is less than value, so we can eliminate it.
The rectangle below and to the right of g is greater than value, so we can eliminate it.
Go to step (2) for each of the two remaining rectangles.
In terms of worst-case complexity, this algorithm does log(M) work to eliminate half the possible solutions, and then recursively calls itself twice on two smaller problems. We do have to repeat a smaller version of that log(M) work for every row, but if the number of rows is small compared to the number of columns, then being able to eliminate all of those columns in logarithmic time starts to become worthwhile.
This gives the algorithm a complexity of T(N,M) = log(M) + 2 * T(M/2, N/2), which Bird shows to be O(N * log(M/N)).
Another approach posted by Craig Gidney describes an algorithm similar the approach above: it examines a row at a time using a step size of M/N. His analysis shows that this results in O(N * log(M/N)) performance as well.
Performance Comparison
Big-O analysis is all well and good, but how well do these approaches work in practice? The chart below examines four algorithms for increasingly "square" arrays:
(The "naive" algorithm simply searches every element of the array. The "recursive" algorithm is described above. The "hybrid" algorithm is an implementation of Gidney's algorithm. For each array size, performance was measured by timing each algorithm over fixed set of 1,000,000 randomly-generated arrays.)
Some notable points:
As expected, the "binary search" algorithms offer the best performance on rectangular arrays and the Saddleback algorithm works the best on square arrays.
The Saddleback algorithm performs worse than the "naive" algorithm for 1-d arrays, presumably because it does multiple comparisons on each item.
The performance hit that the "binary search" algorithms take on square arrays is presumably due to the overhead of running repeated binary searches.
Summary
Clever use of binary search can provide O(N * log(M/N) performance for both rectangular and square arrays. The O(N + M) "saddleback" algorithm is much simpler, but suffers from performance degradation as arrays become increasingly rectangular.
This problem takes Θ(b lg(t)) time, where b = min(w,h) and t=b/max(w,h). I discuss the solution in this blog post.
Lower bound
An adversary can force an algorithm to make Ω(b lg(t)) queries, by restricting itself to the main diagonal:
Legend: white cells are smaller items, gray cells are larger items, yellow cells are smaller-or-equal items and orange cells are larger-or-equal items. The adversary forces the solution to be whichever yellow or orange cell the algorithm queries last.
Notice that there are b independent sorted lists of size t, requiring Ω(b lg(t)) queries to completely eliminate.
Algorithm
(Assume without loss of generality that w >= h)
Compare the target item against the cell t to the left of the top right corner of the valid area
If the cell's item matches, return the current position.
If the cell's item is less than the target item, eliminate the remaining t cells in the row with a binary search. If a matching item is found while doing this, return with its position.
Otherwise the cell's item is more than the target item, eliminating t short columns.
If there's no valid area left, return failure
Goto step 2
Finding an item:
Determining an item doesn't exist:
Legend: white cells are smaller items, gray cells are larger items, and the green cell is an equal item.
Analysis
There are b*t short columns to eliminate. There are b long rows to eliminate. Eliminating a long row costs O(lg(t)) time. Eliminating t short columns costs O(1) time.
In the worst case we'll have to eliminate every column and every row, taking time O(lg(t)*b + b*t*1/t) = O(b lg(t)).
Note that I'm assuming lg clamps to a result above 1 (i.e. lg(x) = log_2(max(2,x))). That's why when w=h, meaning t=1, we get the expected bound of O(b lg(1)) = O(b) = O(w+h).
Code
public static Tuple<int, int> TryFindItemInSortedMatrix<T>(this IReadOnlyList<IReadOnlyList<T>> grid, T item, IComparer<T> comparer = null) {
if (grid == null) throw new ArgumentNullException("grid");
comparer = comparer ?? Comparer<T>.Default;
// check size
var width = grid.Count;
if (width == 0) return null;
var height = grid[0].Count;
if (height < width) {
var result = grid.LazyTranspose().TryFindItemInSortedMatrix(item, comparer);
if (result == null) return null;
return Tuple.Create(result.Item2, result.Item1);
}
// search
var minCol = 0;
var maxRow = height - 1;
var t = height / width;
while (minCol < width && maxRow >= 0) {
// query the item in the minimum column, t above the maximum row
var luckyRow = Math.Max(maxRow - t, 0);
var cmpItemVsLucky = comparer.Compare(item, grid[minCol][luckyRow]);
if (cmpItemVsLucky == 0) return Tuple.Create(minCol, luckyRow);
// did we eliminate t rows from the bottom?
if (cmpItemVsLucky < 0) {
maxRow = luckyRow - 1;
continue;
}
// we eliminated most of the current minimum column
// spend lg(t) time eliminating rest of column
var minRowInCol = luckyRow + 1;
var maxRowInCol = maxRow;
while (minRowInCol <= maxRowInCol) {
var mid = minRowInCol + (maxRowInCol - minRowInCol + 1) / 2;
var cmpItemVsMid = comparer.Compare(item, grid[minCol][mid]);
if (cmpItemVsMid == 0) return Tuple.Create(minCol, mid);
if (cmpItemVsMid > 0) {
minRowInCol = mid + 1;
} else {
maxRowInCol = mid - 1;
maxRow = mid - 1;
}
}
minCol += 1;
}
return null;
}
I would use the divide-and-conquer strategy for this problem, similar to what you suggested, but the details are a bit different.
This will be a recursive search on subranges of the matrix.
At each step, pick an element in the middle of the range. If the value found is what you are seeking, then you're done.
Otherwise, if the value found is less than the value that you are seeking, then you know that it is not in the quadrant above and to the left of your current position. So recursively search the two subranges: everything (exclusively) below the current position, and everything (exclusively) to the right that is at or above the current position.
Otherwise, (the value found is greater than the value that you are seeking) you know that it is not in the quadrant below and to the right of your current position. So recursively search the two subranges: everything (exclusively) to the left of the current position, and everything (exclusively) above the current position that is on the current column or a column to the right.
And ba-da-bing, you found it.
Note that each recursive call only deals with the current subrange only, not (for example) ALL rows above the current position. Just those in the current subrange.
Here's some pseudocode for you:
bool numberSearch(int[][] arr, int value, int minX, int maxX, int minY, int maxY)
if (minX == maxX and minY == maxY and arr[minX,minY] != value)
return false
if (arr[minX,minY] > value) return false; // Early exits if the value can't be in
if (arr[maxX,maxY] < value) return false; // this subrange at all.
int nextX = (minX + maxX) / 2
int nextY = (minY + maxY) / 2
if (arr[nextX,nextY] == value)
{
print nextX,nextY
return true
}
else if (arr[nextX,nextY] < value)
{
if (numberSearch(arr, value, minX, maxX, nextY + 1, maxY))
return true
return numberSearch(arr, value, nextX + 1, maxX, minY, nextY)
}
else
{
if (numberSearch(arr, value, minX, nextX - 1, minY, maxY))
return true
reutrn numberSearch(arr, value, nextX, maxX, minY, nextY)
}
The two main answers give so far seem to be the arguably O(log N) "ZigZag method" and the O(N+M) Binary Search method. I thought I'd do some testing comparing the two methods with some various setups. Here are the details:
The array is N x N square in every test, with N varying from 125 to 8000 (the largest my JVM heap could handle). For each array size, I picked a random place in the array to put a single 2. I then put a 3 everywhere possible (to the right and below of the 2) and then filled the rest of the array with 1. Some of the earlier commenters seemed to think this type of setup would yield worst case run time for both algorithms. For each array size, I picked 100 different random locations for the 2 (search target) and ran the test. I recorded avg run time and worst case run time for each algorithm. Because it was happening too fast to get good ms readings in Java, and because I don't trust Java's nanoTime(), I repeated each test 1000 times just to add a uniform bias factor to all the times. Here are the results:
ZigZag beat binary in every test for both avg and worst case times, however, they are all within an order of magnitude of each other more or less.
Here is the Java code:
public class SearchSortedArray2D {
static boolean findZigZag(int[][] a, int t) {
int i = 0;
int j = a.length - 1;
while (i <= a.length - 1 && j >= 0) {
if (a[i][j] == t) return true;
else if (a[i][j] < t) i++;
else j--;
}
return false;
}
static boolean findBinarySearch(int[][] a, int t) {
return findBinarySearch(a, t, 0, 0, a.length - 1, a.length - 1);
}
static boolean findBinarySearch(int[][] a, int t,
int r1, int c1, int r2, int c2) {
if (r1 > r2 || c1 > c2) return false;
if (r1 == r2 && c1 == c2 && a[r1][c1] != t) return false;
if (a[r1][c1] > t) return false;
if (a[r2][c2] < t) return false;
int rm = (r1 + r2) / 2;
int cm = (c1 + c2) / 2;
if (a[rm][cm] == t) return true;
else if (a[rm][cm] > t) {
boolean b1 = findBinarySearch(a, t, r1, c1, r2, cm - 1);
boolean b2 = findBinarySearch(a, t, r1, cm, rm - 1, c2);
return (b1 || b2);
} else {
boolean b1 = findBinarySearch(a, t, r1, cm + 1, rm, c2);
boolean b2 = findBinarySearch(a, t, rm + 1, c1, r2, c2);
return (b1 || b2);
}
}
static void randomizeArray(int[][] a, int N) {
int ri = (int) (Math.random() * N);
int rj = (int) (Math.random() * N);
a[ri][rj] = 2;
for (int i = 0; i < N; i++) {
for (int j = 0; j < N; j++) {
if (i == ri && j == rj) continue;
else if (i > ri || j > rj) a[i][j] = 3;
else a[i][j] = 1;
}
}
}
public static void main(String[] args) {
int N = 8000;
int[][] a = new int[N][N];
int randoms = 100;
int repeats = 1000;
long start, end, duration;
long zigMin = Integer.MAX_VALUE, zigMax = Integer.MIN_VALUE;
long binMin = Integer.MAX_VALUE, binMax = Integer.MIN_VALUE;
long zigSum = 0, zigAvg;
long binSum = 0, binAvg;
for (int k = 0; k < randoms; k++) {
randomizeArray(a, N);
start = System.currentTimeMillis();
for (int i = 0; i < repeats; i++) findZigZag(a, 2);
end = System.currentTimeMillis();
duration = end - start;
zigSum += duration;
zigMin = Math.min(zigMin, duration);
zigMax = Math.max(zigMax, duration);
start = System.currentTimeMillis();
for (int i = 0; i < repeats; i++) findBinarySearch(a, 2);
end = System.currentTimeMillis();
duration = end - start;
binSum += duration;
binMin = Math.min(binMin, duration);
binMax = Math.max(binMax, duration);
}
zigAvg = zigSum / randoms;
binAvg = binSum / randoms;
System.out.println(findZigZag(a, 2) ?
"Found via zigzag method. " : "ERROR. ");
//System.out.println("min search time: " + zigMin + "ms");
System.out.println("max search time: " + zigMax + "ms");
System.out.println("avg search time: " + zigAvg + "ms");
System.out.println();
System.out.println(findBinarySearch(a, 2) ?
"Found via binary search method. " : "ERROR. ");
//System.out.println("min search time: " + binMin + "ms");
System.out.println("max search time: " + binMax + "ms");
System.out.println("avg search time: " + binAvg + "ms");
}
}
This is a short proof of the lower bound on the problem.
You cannot do it better than linear time (in terms of array dimensions, not the number of elements). In the array below, each of the elements marked as * can be either 5 or 6 (independently of other ones). So if your target value is 6 (or 5) the algorithm needs to examine all of them.
1 2 3 4 *
2 3 4 * 7
3 4 * 7 8
4 * 7 8 9
* 7 8 9 10
Of course this expands to bigger arrays as well. This means that this answer is optimal.
Update: As pointed out by Jeffrey L Whitledge, it is only optimal as the asymptotic lower bound on running time vs input data size (treated as a single variable). Running time treated as two-variable function on both array dimensions can be improved.
I think Here is the answer and it works for any kind of sorted matrix
bool findNum(int arr[][ARR_MAX],int xmin, int xmax, int ymin,int ymax,int key)
{
if (xmin > xmax || ymin > ymax || xmax < xmin || ymax < ymin) return false;
if ((xmin == xmax) && (ymin == ymax) && (arr[xmin][ymin] != key)) return false;
if (arr[xmin][ymin] > key || arr[xmax][ymax] < key) return false;
if (arr[xmin][ymin] == key || arr[xmax][ymax] == key) return true;
int xnew = (xmin + xmax)/2;
int ynew = (ymin + ymax)/2;
if (arr[xnew][ynew] == key) return true;
if (arr[xnew][ynew] < key)
{
if (findNum(arr,xnew+1,xmax,ymin,ymax,key))
return true;
return (findNum(arr,xmin,xmax,ynew+1,ymax,key));
} else {
if (findNum(arr,xmin,xnew-1,ymin,ymax,key))
return true;
return (findNum(arr,xmin,xmax,ymin,ynew-1,key));
}
}
Interesting question. Consider this idea - create one boundary where all the numbers are greater than your target and another where all the numbers are less than your target. If anything is left in between the two, that's your target.
If I'm looking for 3 in your example, I read across the first row until I hit 4, then look for the smallest adjacent number (including diagonals) greater than 3:
1 2 4 5 6
2 3 5 7 8
4 6 8 9 10
5 8 9 10 11
Now I do the same for those numbers less than 3:
1 2 4 5 6
2 3 5 7 8
4 6 8 9 10
5 8 9 10 11
Now I ask, is anything inside the two boundaries? If yes, it must be 3. If no, then there is no 3. Sort of indirect since I don't actually find the number, I just deduce that it must be there. This has the added bonus of counting ALL the 3's.
I tried this on some examples and it seems to work OK.
Binary search through the diagonal of the array is the best option.
We can find out whether the element is less than or equal to the elements in the diagonal.
I've been asking this question in interviews for the better part of a decade and I think there's only been one person who has been able to come up with an optimal algorithm.
My solution has always been:
Binary search the middle diagonal, which is the diagonal running down and right, containing the item at (rows.count/2, columns.count/2).
If the target number is found, return true.
Otherwise, two numbers (u and v) will have been found such that u is smaller than the target, v is larger than the target, and v is one right and one down from u.
Recursively search the sub-matrix to the right of u and top of v and the one to the bottom of u and left of v.
I believe this is a strict improvement over the algorithm given by Nate here, since searching the diagonal often allows a reduction of over half the search space (if the matrix is close to square), whereas searching a row or column always results in an elimination of exactly half.
Here's the code in (probably not terribly Swifty) Swift:
import Cocoa
class Solution {
func searchMatrix(_ matrix: [[Int]], _ target: Int) -> Bool {
if (matrix.isEmpty || matrix[0].isEmpty) {
return false
}
return _searchMatrix(matrix, 0..<matrix.count, 0..<matrix[0].count, target)
}
func _searchMatrix(_ matrix: [[Int]], _ rows: Range<Int>, _ columns: Range<Int>, _ target: Int) -> Bool {
if (rows.count == 0 || columns.count == 0) {
return false
}
if (rows.count == 1) {
return _binarySearch(matrix, rows.lowerBound, columns, target, true)
}
if (columns.count == 1) {
return _binarySearch(matrix, columns.lowerBound, rows, target, false)
}
var lowerInflection = (-1, -1)
var upperInflection = (Int.max, Int.max)
var currentRows = rows
var currentColumns = columns
while (currentRows.count > 0 && currentColumns.count > 0 && upperInflection.0 > lowerInflection.0+1) {
let rowMidpoint = (currentRows.upperBound + currentRows.lowerBound) / 2
let columnMidpoint = (currentColumns.upperBound + currentColumns.lowerBound) / 2
let value = matrix[rowMidpoint][columnMidpoint]
if (value == target) {
return true
}
if (value > target) {
upperInflection = (rowMidpoint, columnMidpoint)
currentRows = currentRows.lowerBound..<rowMidpoint
currentColumns = currentColumns.lowerBound..<columnMidpoint
} else {
lowerInflection = (rowMidpoint, columnMidpoint)
currentRows = rowMidpoint+1..<currentRows.upperBound
currentColumns = columnMidpoint+1..<currentColumns.upperBound
}
}
if (lowerInflection.0 == -1) {
lowerInflection = (upperInflection.0-1, upperInflection.1-1)
} else if (upperInflection.0 == Int.max) {
upperInflection = (lowerInflection.0+1, lowerInflection.1+1)
}
return _searchMatrix(matrix, rows.lowerBound..<lowerInflection.0+1, upperInflection.1..<columns.upperBound, target) || _searchMatrix(matrix, upperInflection.0..<rows.upperBound, columns.lowerBound..<lowerInflection.1+1, target)
}
func _binarySearch(_ matrix: [[Int]], _ rowOrColumn: Int, _ range: Range<Int>, _ target: Int, _ searchRow : Bool) -> Bool {
if (range.isEmpty) {
return false
}
let midpoint = (range.upperBound + range.lowerBound) / 2
let value = (searchRow ? matrix[rowOrColumn][midpoint] : matrix[midpoint][rowOrColumn])
if (value == target) {
return true
}
if (value > target) {
return _binarySearch(matrix, rowOrColumn, range.lowerBound..<midpoint, target, searchRow)
} else {
return _binarySearch(matrix, rowOrColumn, midpoint+1..<range.upperBound, target, searchRow)
}
}
}
A. Do a binary search on those lines where the target number might be on.
B. Make it a graph : Look for the number by taking always the smallest unvisited neighbour node and backtracking when a too big number is found
Binary search would be the best approach, imo. Starting at 1/2 x, 1/2 y will cut it in half. IE a 5x5 square would be something like x == 2 / y == 3 . I rounded one value down and one value up to better zone in on the direction of the targeted value.
For clarity the next iteration would give you something like x == 1 / y == 2 OR x == 3 / y == 5
Well, to begin with, let us assume we are using a square.
1 2 3
2 3 4
3 4 5
1. Searching a square
I would use a binary search on the diagonal. The goal is the locate the smaller number that is not strictly lower than the target number.
Say I am looking for 4 for example, then I would end up locating 5 at (2,2).
Then, I am assured that if 4 is in the table, it is at a position either (x,2) or (2,x) with x in [0,2]. Well, that's just 2 binary searches.
The complexity is not daunting: O(log(N)) (3 binary searches on ranges of length N)
2. Searching a rectangle, naive approach
Of course, it gets a bit more complicated when N and M differ (with a rectangle), consider this degenerate case:
1 2 3 4 5 6 7 8
2 3 4 5 6 7 8 9
10 11 12 13 14 15 16 17
And let's say I am looking for 9... The diagonal approach is still good, but the definition of diagonal changes. Here my diagonal is [1, (5 or 6), 17]. Let's say I picked up [1,5,17], then I know that if 9 is in the table it is either in the subpart:
5 6 7 8
6 7 8 9
10 11 12 13 14 15 16
This gives us 2 rectangles:
5 6 7 8 10 11 12 13 14 15 16
6 7 8 9
So we can recurse! probably beginning by the one with less elements (though in this case it kills us).
I should point that if one of the dimensions is less than 3, we cannot apply the diagonal methods and must use a binary search. Here it would mean:
Apply binary search on 10 11 12 13 14 15 16, not found
Apply binary search on 5 6 7 8, not found
Apply binary search on 6 7 8 9, not found
It's tricky because to get good performance you might want to differentiate between several cases, depending on the general shape....
3. Searching a rectangle, brutal approach
It would be much easier if we dealt with a square... so let's just square things up.
1 2 3 4 5 6 7 8
2 3 4 5 6 7 8 9
10 11 12 13 14 15 16 17
17 . . . . . . 17
. .
. .
. .
17 . . . . . . 17
We now have a square.
Of course, we will probably NOT actually create those rows, we could simply emulate them.
def get(x,y):
if x < N and y < M: return table[x][y]
else: return table[N-1][M-1] # the max
so it behaves like a square without occupying more memory (at the cost of speed, probably, depending on cache... oh well :p)
EDIT:
I misunderstood the question. As the comments point out this only works in the more restricted case.
In a language like C that stores data in row-major order, simply treat it as a 1D array of size n * m and use a binary search.
I have a recursive Divide & Conquer Solution.
Basic Idea for one step is: We know that the Left-Upper(LU) is smallest and the right-bottom(RB) is the largest no., so the given No(N) must: N>=LU and N<=RB
IF N==LU and N==RB::::Element Found and Abort returning the position/Index
If N>=LU and N<=RB = FALSE, No is not there and abort.
If N>=LU and N<=RB = TRUE, Divide the 2D array in 4 equal parts of 2D array each in logical manner..
And then apply the same algo step to all four sub-array.
My Algo is Correct I have implemented on my friends PC.
Complexity: each 4 comparisons can b used to deduce the total no of elements to one-fourth at its worst case..
So My complexity comes to be 1 + 4 x lg(n) + 4
But really expected this to be working on O(n)
I think something is wrong somewhere in my calculation of Complexity, please correct if so..
The optimal solution is to start at the top-left corner, that has minimal value. Move diagonally downwards to the right until you hit an element whose value >= value of the given element. If the element's value is equal to that of the given element, return found as true.
Otherwise, from here we can proceed in two ways.
Strategy 1:
Move up in the column and search for the given element until we reach the end. If found, return found as true
Move left in the row and search for the given element until we reach the end. If found, return found as true
return found as false
Strategy 2:
Let i denote the row index and j denote the column index of the diagonal element we have stopped at. (Here, we have i = j, BTW). Let k = 1.
Repeat the below steps until i-k >= 0
Search if a[i-k][j] is equal to the given element. if yes, return found as true.
Search if a[i][j-k] is equal to the given element. if yes, return found as true.
Increment k
1 2 4 5 6
2 3 5 7 8
4 6 8 9 10
5 8 9 10 11
public boolean searchSortedMatrix(int arr[][] , int key , int minX , int maxX , int minY , int maxY){
// base case for recursion
if(minX > maxX || minY > maxY)
return false ;
// early fails
// array not properly intialized
if(arr==null || arr.length==0)
return false ;
// arr[0][0]> key return false
if(arr[minX][minY]>key)
return false ;
// arr[maxX][maxY]<key return false
if(arr[maxX][maxY]<key)
return false ;
//int temp1 = minX ;
//int temp2 = minY ;
int midX = (minX+maxX)/2 ;
//if(temp1==midX){midX+=1 ;}
int midY = (minY+maxY)/2 ;
//if(temp2==midY){midY+=1 ;}
// arr[midX][midY] = key ? then value found
if(arr[midX][midY] == key)
return true ;
// alas ! i have to keep looking
// arr[midX][midY] < key ? search right quad and bottom matrix ;
if(arr[midX][midY] < key){
if( searchSortedMatrix(arr ,key , minX,maxX , midY+1 , maxY))
return true ;
// search bottom half of matrix
if( searchSortedMatrix(arr ,key , midX+1,maxX , minY , maxY))
return true ;
}
// arr[midX][midY] > key ? search left quad matrix ;
else {
return(searchSortedMatrix(arr , key , minX,midX-1,minY,midY-1));
}
return false ;
}
I suggest, store all characters in a 2D list. then find index of required element if it exists in list.
If not present print appropriate message else print row and column as:
row = (index/total_columns) and column = (index%total_columns -1)
This will incur only the binary search time in a list.
Please suggest any corrections. :)
If O(M log(N)) solution is ok for an MxN array -
template <size_t n>
struct MN * get(int a[][n], int k, int M, int N){
struct MN *result = new MN;
result->m = -1;
result->n = -1;
/* Do a binary search on each row since rows (and columns too) are sorted. */
for(int i = 0; i < M; i++){
int lo = 0; int hi = N - 1;
while(lo <= hi){
int mid = lo + (hi-lo)/2;
if(k < a[i][mid]) hi = mid - 1;
else if (k > a[i][mid]) lo = mid + 1;
else{
result->m = i;
result->n = mid;
return result;
}
}
}
return result;
}
Working C++ demo.
Please do let me know if this wouldn't work or if there is a bug it it.
class Solution {
public boolean searchMatrix(int[][] matrix, int target) {
if(matrix == null)
return false;
int i=0;
int j=0;
int m = matrix.length;
int n = matrix[0].length;
boolean found = false;
while(i<m && !found){
while(j<n && !found){
if(matrix[i][j] == target)
found = true;
if(matrix[i][j] < target)
j++;
else
break;
}
i++;
j=0;
}
return found;
}}
129 / 129 test cases passed.
Status: Accepted
Runtime: 39 ms
Memory Usage: 55 MB
Given a square matrix as follows:
[ a b c ]
[ d e f ]
[ i j k ]
We know that a < c, d < f, i < k. What we don't know is whether d < c or d > c, etc. We have guarantees only in 1-dimension.
Looking at the end elements (c,f,k), we can do a sort of filter: is N < c ? search() : next(). Thus, we have n iterations over the rows, with each row taking either O( log( n ) ) for binary search or O( 1 ) if filtered out.
Let me given an EXAMPLE where N = j,
1) Check row 1. j < c? (no, go next)
2) Check row 2. j < f? (yes, bin search gets nothing)
3) Check row 3. j < k? (yes, bin search finds it)
Try again with N = q,
1) Check row 1. q < c? (no, go next)
2) Check row 2. q < f? (no, go next)
3) Check row 3. q < k? (no, go next)
There is probably a better solution out there but this is easy to explain.. :)
As this is an interview question, it would seem to lead towards a discussion of Parallel programming and Map-reduce algorithms.
See http://code.google.com/intl/de/edu/parallel/mapreduce-tutorial.html