Iterative segment tree implmentation - c++

I'm referring to this post :
https://www.geeksforgeeks.org/segment-tree-efficient-implementation/
Pasting the code here for reference. My questions are below the code.
#include <bits/stdc++.h>
using namespace std;
// limit for array size
const int N = 100000;
int n; // array size
// Max size of tree
int tree[2 * N];
// function to build the tree
void build( int arr[])
{
// insert leaf nodes in tree
for (int i=0; i<n; i++)
tree[n+i] = arr[i];
// build the tree by calculating parents
for (int i = n - 1; i > 0; --i)
tree[i] = tree[i<<1] + tree[i<<1 | 1];
}
// function to update a tree node
void updateTreeNode(int p, int value)
{
// set value at position p
tree[p+n] = value; // {Question} Why is this assigned before updating the parent index to access?
p = p+n;
// move upward and update parents
for (int i=p; i > 1; i >>= 1)
tree[i>>1] = tree[i] + tree[i^1];
}
// function to get sum on interval [l, r)
int query(int l, int r)
{
int res = 0;
// loop to find the sum in the range
for (l += n, r += n; l < r; l >>= 1, r >>= 1)
{
if (l&1)
res += tree[l++];
if (r&1)
res += tree[--r];
// {Question} What happens if !(l&1) or !(r&1) ? res is not accumulated anywhere. Isn't this wrong?
}
return res;
}
// driver program to test the above function
int main()
{
int a[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12};
// n is global
n = sizeof(a)/sizeof(a[0]);
// build tree
build(a);
// print the sum in range(1,2) index-based
cout << query(1, 3)<<endl;
// modify element at 2nd index
updateTreeNode(2, 1);
// print the sum in range(1,2) index-based
cout << query(1, 3)<<endl;
return 0;
}
Questions:
Within updateTreeNode where value is being set to parent,
why is the parent being incremented by array size after assigning the value in the tree?
Shouldn't it be before?
The original line of code this post has implemented this from reads like this:
for (tree[parent += n] = value; parent > 1; parent >>= 1)
where parent=parent+n executes first.
Can someone help me understand why this code is still functioning properly?
Within query which returns a sum in the interval [l, r), this code seems to add
result only for odd values of l.
Yet, I see that the result correctly returns the sum for even-valued-intervals.
The result should be skipping accumulating even-valued intervals since there is no else <accumulate result>, right? What am I missing?

Question 1
The variable p doesn't mean parent, it means child. In for loop, i is child node, and we update value of i's parent.
tree[p+n] = value;: update the value of leave node (node without children). Then we update value of node's parent from the leave node. tree[i>>1] = tree[i] + tree[i^1];, tree[i>>1] is tree[i]' parent.
For example: the array size is 16 (the tree size is 32) and I want to update arr[8]. So I call updateTreeNode(8, value). First three[8+16] is updated, which corresponds to arr[8]. Then p is set to 24. In for loop, we update p's parent (tree[12]), then set p to p/2 (p=12), until p doesn't have parent.
Qustion 2
If l and r is even, we add the parent's value in next recurrence. That's the function of segment tree, to avoid querying each element in the interval.
For example: the array size is 16 and I want to query [8,10). In segment tree, the interval is [24,26). We don't need to add value of tree[24] and tree[25], we add value of tree[12]!

Related

Finding the number of sub arrays that have a sum of K

I am trying to find the number of sub arrays that have a sum equal to k:
int subarraySum(vector<int>& nums, int k)
{
int start, end, curr_sum = 0, count = 0;
start = 0, end = 0;
while (end < (int)nums.size())
{
curr_sum = curr_sum + nums[end];
end++;
while (start < end && curr_sum >= k)
{
if (curr_sum == k)
count++;
curr_sum = curr_sum - nums[start];
start++;
}
}
return count;
}
The above code I have written, works for most cases, but fails for the following:
array = {-1, -1, 1} with k = 0
I have tried to add another while loop to iterate from the start and go up the array until it reaches the end:
int subarraySum(vector<int>& nums, int k)
{
int start, end, curr_sum = 0, count = 0;
start = 0, end = 0;
while (end < (int)nums.size())
{
curr_sum = curr_sum + nums[end];
end++;
while (start < end && curr_sum >= k)
{
if (curr_sum == k)
count++;
curr_sum = curr_sum - nums[start];
start++;
}
}
while (start < end)
{
if (curr_sum == k)
count++;
curr_sum = curr_sum - nums[start];
start++;
}
return count;
}
Why is this not working? I am sliding the window until the last element is reached, which should have found a sum equal to k? How can I solve this issue?
Unfortunately, you did not program a sliding window in the correct way. And a sliding window is not really a solution for this problem. One of your main issues is, that you do not move the start of the window based on the proper conditions. You always sum up and wait until the sum is greater than the search value.
This will not really work. Especially for your example -1, -1, 1. The running sum of this is: -1, -2, -1 and you do not see the 0, although it is there. You may have the idea to write while (start < end && curr_sum != k), but this will also not work, because you handle the start pointer not correctly.
Your approach will lead to the brute force solution that typically takes something like N*N loop operations, where N is the size of the array. This, because we need a double nested loop.
That will of course always work, but maybe very time-consuming, and, in the end, too slow.
Anyway. Let us implement that. We will start from each value in the std::vector and try out all sub arrays starting from the beginning value. We must evaluate all following values in the std::vector, because for example the last value could be a big negative number and bring down the sum again to the search value.
We could implement this for example like the following:
#include <iostream>
#include <vector>
using namespace std;
int subarraySum(vector<int>& numbers, int searchSumValue) {
// Here we will store the result
int resultingCount{};
// Iterate over all values in the array. So, use all different start values
for (std::size_t i{}; i < numbers.size(); ++i) {
// Here we stor the running sum of the elements in the vector
int sum{ numbers[i] };
// Check for trivial case. A one-element sub-array does already match the search value
if (sum == searchSumValue) ++resultingCount;
// Now we build all subarrays beginning with the start value
for (std::size_t k{ i + 1 }; k < numbers.size(); ++k) {
sum += numbers[k];
if (sum == searchSumValue) ++resultingCount;
}
}
return resultingCount;
}
int main() {
vector v{ -1,-1,1 };
std::cout << subarraySum(v, 0);
}
.
But, as said, the above is often too slow for big vectors and there is indeed a better solution available, which is based on a DP (dynamic programming) algorithm.
It uses so-called prefix sums, running sums, based on the running sum before the current evaluated value.
We need to show an example. Let's use a std::vector with 5 values {1,2,3,4,5}. And we want to look subarrays with a sum of 9.
We can “guess” that there are 2 subarrays: {2,3,4} and {4,5} that have a sum of 9.
Let us investigate further
Index 0 1 2 3 4
Value 1 2 3 4 5
We can now add a running sum and see, how much delta we have between the current evaluated element and the left neighbor or over-next neighbor and so on. And if we have a delta that is equal to our search value, then we must have a subarray building this sum.
Running Sum 1 3 6 10 15
Deltas of 2 3 4 5 against next left
Running sum 5 7 9 against next next left
9 12 against next next next left
Example {2,3,4}. If we evaluate the 4 with a running sum of 10, and subtract the search value 9, then we get the previous running sum 1. “1+9=10” all values are there.
Example {4,5}. If we evaluate the 5 with a running sum of 15, and subtract the search value 9, then we get the previous running sum = 6. “6+9=15” all values are there.
We can find all solutions using the same approach.
So, the only thing we need to do, is to subtract the search value from the current running sum and see, if we have this running sum already calculated before.
Like: “Search-Value” + “previously Calculated Sum” = “Current Running Sum”.
Or: “Current Running Sum” – “Search-Value” = “previously Calculated Sum”
Again, we need to do the subtraction and check, if we already calculated such a sum previously.
So, we need to store all previously calculated running sums. And, because such a sum may appear more than one, we need to find occurrences of equal running sums and count them.
It is very hard to digest, and you need to think a while to understand.
With the above wisdom, you can draft the below potential solution.
#include <iostream>
#include <vector>
#include <unordered_map>
int subarraySum(std::vector<int>& numbers, int searchSumValue) {
// Here we will store the result
int resultingSubarrayCount{};
// Here we will stor all running sums and how ofthen their value appeared
std::unordered_map<int, int> countOfRunningSums;
// Continuosly calculating the running sum
int runningSum{};
// And initialize the first value
countOfRunningSums[runningSum] = 1;
// Now iterate over all values in the vector
for (const int n : numbers) {
// Calculate the running sum
runningSum += n;
// Check, if we have the searched value already available
// And add the number of occurences to our resulting number of subarrays
resultingSubarrayCount += countOfRunningSums[runningSum - searchSumValue];
// Store the new running sum. Respectively. Add 1 to the counter, if the running sum was alreadyy existing
countOfRunningSums[runningSum]++;
}
return resultingSubarrayCount;
}
int main() {
std::vector v{ 1,2,3,4,5 };
std::cout << subarraySum(v, 9);
}

Intuition behind using a monotonic stack

I am solving a question on LeetCode.com:
Given an array of integers A, find the sum of min(B), where B ranges over every (contiguous) subarray of A. Since the answer may be large, return the answer modulo 10^9 + 7.
Input: [3,1,2,4]
Output: 17
Explanation: Subarrays are [3], [1], [2], [4], [3,1], [1,2], [2,4], [3,1,2], [1,2,4], [3,1,2,4]. Minimums are 3, 1, 2, 4, 1, 1, 2, 1, 1, 1. Sum is 17.
A highly upvoted solution is as below:
class Solution {
public:
int sumSubarrayMins(vector<int>& A) {
stack<pair<int, int>> in_stk_p, in_stk_n;
// left is for the distance to previous less element
// right is for the distance to next less element
vector<int> left(A.size()), right(A.size());
//initialize
for(int i = 0; i < A.size(); i++) left[i] = i + 1;
for(int i = 0; i < A.size(); i++) right[i] = A.size() - i;
for(int i = 0; i < A.size(); i++){
// for previous less
while(!in_stk_p.empty() && in_stk_p.top().first > A[i]) in_stk_p.pop();
left[i] = in_stk_p.empty()? i + 1: i - in_stk_p.top().second;
in_stk_p.push({A[i],i});
// for next less
while(!in_stk_n.empty() && in_stk_n.top().first > A[i]){
auto x = in_stk_n.top();in_stk_n.pop();
right[x.second] = i - x.second;
}
in_stk_n.push({A[i], i});
}
int ans = 0, mod = 1e9 +7;
for(int i = 0; i < A.size(); i++){
ans = (ans + A[i]*left[i]*right[i])%mod;
}
return ans;
}
};
My question is: what is the intuition behind using a monotonically increasing stack for this? How does it help calculate the minimums in the various subarrays?
Visualize the array as a line graph, with (local) minima as valleys. Each value is relevant for a range that extends from just after the previous smaller value (if any) to just before the next smaller value (if any). (Even a value larger than its neighbors matters when considering the singleton subarray that contains it.) The variables left and right track that range.
Recognizing that a value shadows every value larger than it in each direction separately, the stack maintains a list of previous, unshadowed minima for two purposes: identifying how far back a new small number’s range extends and (at the same time) how far forward the invalidated minima’s ranges extend. The code uses a separate stack for each purpose, but there’s no need: each has the same contents after every iteration of the (outer) loop.

Modify lazy propogation in segment tree

I recently read about lazy propogation in segment tree and coded it too.But i got stuck when suppose instead of adding value(=val) i need to divide by value.How to do it ?
Please help
My update function is as follow :
void update_tree(int node, int a, int b, int i, int j, int value) {
if(lazy[node] != 0) { // This node needs to be updated
tree[node] += lazy[node]; // Update it
if(a != b) {
lazy[node*2] += lazy[node]; // Mark child as lazy
lazy[node*2+1] += lazy[node]; // Mark child as lazy
}
lazy[node] = 0; // Reset it
}
if(a > b || a > j || b < i) // Current segment is not within range [i, j]
return;
if(a >= i && b <= j) { // Segment is fully within range
tree[node] += value;
if(a != b) { // Not leaf node
lazy[node*2] += value;
lazy[node*2+1] += value;
}
return;
}
update_tree(node*2, a, (a+b)/2, i, j, value); // Updating left child
update_tree(1+node*2, 1+(a+b)/2, b, i, j, value); // Updating right child
tree[node] = max(tree[node*2], tree[node*2+1]); // Updating root with max value
}
HINTS
Suppose you need to divide by a fixed value of K.
One possibility would be to convert your numbers to base K and in each node maintain an array of numbers A[], where A[i] is the total in all lower nodes of all digits in position i (when thought of as a base K number).
So, for example, if K was 10, then A[0] would store the total of all the units, while A[1] would store the total of all the tens.
The reason to do this is that it then becomes easy to divide lazily by K, all you need to do is set A[i]=A[i+1] and you can use the same lazy update trick as in your code.
EXAMPLE
Suppose we had an array 5,11,20,100 and K was 10
We would construct a node for element 5,11 containing the value:
Total = A[1]*10+A[0]*1 with A[1]=1 and A[0]=5+1 (the sum of the unit values)
we would also have a node for 20,100 containing the value:
Total = A[2]*100+A[1]*10+A[0]*1 with A[2]=1,A[1]=2,A[0]=0
and a node for the entire 5,11,20,100 array with:
Total = A[2]*100+A[1]*10+A[0]*1 with A[2]=1,A[1]=2+1,A[0]=5+1
If we then wanted to divide the whole array by 10, we would simply change the array elements for the top node:
A=[1,3,6] changes to [0,1,3]
and then we could query the sum of all the node by computing:
Total = A[2]*100+A[1]*10+A[0]*1 = 0*100+1*10+3*1=13
which is the same as
(5/10=0)+(11/10=1)+(20/10=2)+(100/10=10)

Maintain a sorted array in O(1)?

We have a sorted array and we would like to increase the value of one index by only 1 unit (array[i]++), such that the resulting array is still sorted. Is this possible in O(1)?
It is fine to use any data structure possible in STL and C++.
In a more specific case, if the array is initialised by all 0 values, and it is always incrementally constructed only by increasing a value of an index by one, is there an O(1) solution?
I haven't worked this out completely, but I think the general idea might help for integers at least. At the cost of more memory, you can maintain a separate data-structure that maintains the ending index of a run of repeated values (since you want to swap your incremented value with the ending index of the repeated value). This is because it's with repeated values that you run into the worst case O(n) runtime: let's say you have [0, 0, 0, 0] and you increment the value at location 0. Then it is O(n) to find out the last location (3).
But let's say that you maintain the data-structure I mentioned (a map would works because it has O(1) lookup). In that case you would have something like this:
0 -> 3
So you have a run of 0 values that end at location 3. When you increment a value, let's say at location i, you check to see if the new value is greater than the value at i + 1. If it is not, you are fine. But if it is, you look to see if there is an entry for this value in the secondary data-structure. If there isn't, you can simply swap. If there is an entry, you look up the ending-index and then swap with the value at that location. You then make any changes you need to the secondary data-structure to reflect the new state of the array.
A more thorough example:
[0, 2, 3, 3, 3, 4, 4, 5, 5, 5, 7]
The secondary data-structure is:
3 -> 4
4 -> 6
5 -> 9
Let's say you increment the value at location 2. So you have incremented 3, to 4. The array now looks like this:
[0, 2, 4, 3, 3, 4, 4, 5, 5, 5, 7]
You look at the next element, which is 3. You then look up the entry for that element in the secondary data-structure. The entry is 4, which means that there is a run of 3's that end at 4. This means that you can swap the value from the current location with the value at index 4:
[0, 2, 3, 3, 4, 4, 4, 5, 5, 5, 7]
Now you will also need to update the secondary data-structure. Specifically, there the run of 3's ends one index early, so you need to decrement that value:
3 -> 3
4 -> 6
5 -> 9
Another check you will need to do is to see if the value is repeated anymore. You can check that by looking at the i - 1th and the i + 1th locations to see if they are the same as the value in question. If neither are equal, then you can remove the entry for this value from the map.
Again, this is just a general idea. I will have to code it out to see if it works out the way I thought about it.
Please feel free to poke holes.
UPDATE
I have an implementation of this algorithm here in JavaScript. I used JavaScript just so I could do it quickly. Also, because I coded it up pretty quickly it can probably be cleaned up. I do have comments though. I'm not doing anything esoteric either, so this should be easily portable to C++.
There are essentially two parts to the algorithm: the incrementing and swapping (if necessary), and book-keeping done on the map that keeps track of our ending indices for runs of repeated values.
The code contains a testing harness that starts with an array of zeroes and increments random locations. At the end of every iteration, there is a test to ensure that the array is sorted.
var array = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0];
var endingIndices = {0: 9};
var increments = 10000;
for(var i = 0; i < increments; i++) {
var index = Math.floor(Math.random() * array.length);
var oldValue = array[index];
var newValue = ++array[index];
if(index == (array.length - 1)) {
//Incremented element is the last element.
//We don't need to swap, but we need to see if we modified a run (if one exists)
if(endingIndices[oldValue]) {
endingIndices[oldValue]--;
}
} else if(index >= 0) {
//Incremented element is not the last element; it is in the middle of
//the array, possibly even the first element
var nextIndexValue = array[index + 1];
if(newValue === nextIndexValue) {
//If the new value is the same as the next value, we don't need to swap anything. But
//we are doing some book-keeping later with the endingIndices map. That code requires
//the ending index (i.e., where we moved the incremented value to). Since we didn't
//move it anywhere, the endingIndex is simply the index of the incremented element.
endingIndex = index;
} else if(newValue > nextIndexValue) {
//If the new value is greater than the next value, we will have to swap it
var swapIndex = -1;
if(!endingIndices[nextIndexValue]) {
//If the next value doesn't have a run, then location we have to swap with
//is just the next index
swapIndex = index + 1;
} else {
//If the next value has a run, we get the swap index from the map
swapIndex = endingIndices[nextIndexValue];
}
array[index] = nextIndexValue;
array[swapIndex] = newValue;
endingIndex = swapIndex;
} else {
//If the next value is already greater, there is nothing we need to swap but we do
//need to do some book-keeping with the endingIndices map later, because it is
//possible that we modified a run (the value might be the same as the value that
//came before it). Since we don't have anything to swap, the endingIndex is
//effectively the index that we are incrementing.
endingIndex = index;
}
//Moving the new value to its new position may have created a new run, so we need to
//check for that. This will only happen if the new position is not at the end of
//the array, and the new value does not have an entry in the map, and the value
//at the position after the new position is the same as the new value
if(endingIndex < (array.length - 1) &&
!endingIndices[newValue] &&
array[endingIndex + 1] == newValue) {
endingIndices[newValue] = endingIndex + 1;
}
//We also need to check to see if the old value had an entry in the
//map because now that run has been shortened by one.
if(endingIndices[oldValue]) {
var newEndingIndex = --endingIndices[oldValue];
if(newEndingIndex == 0 ||
(newEndingIndex > 0 && array[newEndingIndex - 1] != oldValue)) {
//In this case we check to see if the old value only has one entry, in
//which case there is no run of values and so we will need to remove
//its entry from the map. This happens when the new ending-index for this
//value is the first location (0) or if the location before the new
//ending-index doesn't contain the old value.
delete endingIndices[oldValue];
}
}
}
//Make sure that the array is sorted
for(var j = 0; j < array.length - 1; j++) {
if(array[j] > array[j + 1]) {
throw "Array not sorted; Value at location " + j + "(" + array[j] + ") is greater than value at location " + (j + 1) + "(" + array[j + 1] + ")";
}
}
}
In a more specific case, if the array is initialised by all 0 values, and it is always incrementally constructed only by increasing a value of an index by one, is there an O(1) solution?
No. Given an array of all 0's: [0, 0, 0, 0, 0]. If you increment the first value, giving [1, 0, 0, 0, 0], then you will have to make 4 swaps to ensure that it remains sorted.
Given a sorted array with no duplicates, then the answer is yes. But after the first operation (i.e. the first time you increment), then you could potentially have duplicates. The more increments you do, the higher the likelihood is that you'll have duplicates, and the more likely it'll take O(n) to keep that array sorted.
If all you have is the array, it's impossible to guarantee less than O(n) time per increment. If what you're looking for is a data structure that supports sorted order and lookup by index, then you probably want an order stastic tree.
If the values are small, counting sort will work. Represent the array [0,0,0,0] as {4}. Incrementing any zero gives {3,1} : 3 zeroes and a one. In general, to increment any value x, deduct one from the count of x and increment the count of {x+1}. The space efficiency is O(N), though, where N is the highest value.
It depends on how many items can have the same value. If more items can have the same value, then it is not possible to have O(1) with ordinary arrays.
Let's do an example: suppose array[5] = 21, and you want to do array[5]++:
Increment the item:
array[5]++
(which is O(1) because it is an array).
So, now array[5] = 22.
Check the next item (i.e., array[6]):
If array[6] == 21, then you have to keep checking new items (i.e., array[7] and so on) until you find a value higher than 21. At that point you can swap the values. This search is not O(1) because potentially you have to scan the whole array.
Instead, if items cannot have the same value, then you have:
Increment the item:
array[5]++
(which is O(1) because it is an array).
So, now array[5] = 22.
The next item cannot be 21 (because two items cannot have the same value), so it must have a value > 21 and the array is already sorted.
So you take sorted array and hashtable. You go over array to figure out 'flat' areas - where elements are of the same value. For every flat area you have to figure out three things 1) where it starts (index of first element) 2) what is it's value 3) what is the value of next element (the next bigger). Then put this tuple into the hashtable, where the key will be element value. This is prerequisite and it's complexity doesn't really matter.
Then when you increase some element (index i) you look up a table for index of next bigger element (call it j), and swap i with i - 1. Then 1) add new entry to hashtable 2) update existing entry for it's previous value.
With perfect hashtable (or limited range of possible values) it will be almost O(1). The downside: it will not be stable.
Here is some code:
#include <iostream>
#include <unordered_map>
#include <vector>
struct Range {
int start, value, next;
};
void print_ht(std::unordered_map<int, Range>& ht)
{
for (auto i = ht.begin(); i != ht.end(); i++) {
Range& r = (*i).second;
std::cout << '(' << r.start << ", "<< r.value << ", "<< r.next << ") ";
}
std::cout << std::endl;
}
void increment_el(int i, std::vector<int>& array, std::unordered_map<int, Range>& ht)
{
int val = array[i];
array[i]++;
//Pick next bigger element
Range& r = ht[val];
//Do the swapping, so last element of that range will be first
std::swap(array[i], array[ht[r.next].start - 1]);
//Update hashtable
ht[r.next].start--;
}
int main(int argc, const char * argv[])
{
std::vector<int> array = {1, 1, 1, 2, 2, 3};
std::unordered_map<int, Range> ht;
int start = 0;
int value = array[0];
//Build indexing hashtable
for (int i = 0; i <= array.size(); i++) {
int cur_value = i < array.size() ? array[i] : -1;
if (cur_value > value || i == array.size()) {
ht[value] = {start, value, cur_value};
start = i;
value = cur_value;
}
}
print_ht(ht);
//Now let's increment first element
increment_el(0, array, ht);
print_ht(ht);
increment_el(3, array, ht);
print_ht(ht);
for (auto i = array.begin(); i != array.end(); i++)
std::cout << *i << " ";
return 0;
}
Yes and no.
Yes if the list contains only unique integers, as that means you only need to check the next value. No in any other situation. If the values are not unique, incrementing the first of N duplicate values means that it must move N positions. If the values are floating-point, you may have thousands of values between x and x+1
It's important to be very clear about the requirements; the simplest way is to express the problem as an ADT (Abstract Datatype), listing the required operations and complexities.
Here's what I think you are looking for: a datatype which provides the following operations:
Construct(n): Create a new object of size n all of whose values are 0.
Value(i): Return the value at index i.
Increment(i): Increment the value at index i.
Least(): Return the index of the element with least value (or one such element if there are several).
Next(i): Return the index of the next element after element i in a sorted traversal starting at Least(), such that the traversal will return every element.
Aside from the Constructor, we want every one of the above operations to have complexity O(1). We also want the object to occupy O(n) space.
The implementation uses a list of buckets; each bucket has a value and a list of elements. Each element has an index, a pointer to the bucket it is part of. Finally, we have an array of pointers to elements. (In C++, I'd probably use iterators rather than pointers; in another language, I'd probably use intrusive lists.) The invariants are that no bucket is ever empty, and the value of the buckets are strictly monotonically increasing.
We start with a single bucket with value 0 which has a list of n elements.
Value(i) is implemented by returning the value of the bucket of the element referenced by the iterator at element i of the array. Least() is the index of the first element in the first bucket. Next(i) is the index of the next element after the one referenced by the iterator at element i, unless that iterator is already pointing at the end of the the list in which case it is the first element in the next bucket, unless the element's bucket is the last bucket, in which case we're at the end of the element list.
The only interface of interest is Increment(i), which is as follows:
If element i is the only element in its bucket (i.e. there is no next element in the bucket list, and element i is the first element in the bucket list):
Increment the value of the associated bucket.
If the next bucket has the same value, append the next bucket's element list to this bucket's element list (this is O(1), regardless of the list's size, because it is just a pointer swap), and then delete the next bucket.
If element i is not the only element in its bucket, then:
Remove it from its bucket list.
If the next bucket has the next sequential value, then push element i onto the next bucket's list.
Otherwise, the next bucket's value is larger, then create a new bucket with the next sequential value and only element i and insert it between this bucket and the next one.
just iterate along the array from the modified element until you find the correct place, then swap. Average case complexity is O(N) where N is the average number of duplicates. Worst case is O(n) where n is the length of the array. As long as N isn't large and doesn't scale badly with n, you're fine and can probably pretend it's O(1) for practical purposes.
If duplicates are the norm and/or scale strongly with n, then there are better solutions, see other responses.
I think that it is possible without using a hashtable. I have an implementation here:
#include <cstdio>
#include <vector>
#include <cassert>
// This code is a solution for http://stackoverflow.com/questions/19957753/maintain-a-sorted-array-in-o1
//
// """We have a sorted array and we would like to increase the value of one index by only 1 unit
// (array[i]++), such that the resulting array is still sorted. Is this possible in O(1)?"""
// The obvious implementation, which has O(n) worst case increment.
class LinearIncrementor
{
public:
LinearIncrementor(int numElems);
int valueAt(int index) const;
void incrementAt(int index);
private:
std::vector<int> m_values;
};
// Free list to store runs of same values
class RunList
{
public:
struct Run
{
int m_end; // end index of run, inclusive, or next object in free list
int m_value; // value at this run
};
RunList();
int allocateRun(int endIndex, int value);
void freeRun(int index);
Run& runAt(int index);
const Run& runAt(int index) const;
private:
std::vector<Run> m_runs;
int m_firstFree;
};
// More optimal implementation, which increments in O(1) time
class ConstantIncrementor
{
public:
ConstantIncrementor(int numElems);
int valueAt(int index) const;
void incrementAt(int index);
private:
std::vector<int> m_runIndices;
RunList m_runs;
};
LinearIncrementor::LinearIncrementor(int numElems)
: m_values(numElems, 0)
{
}
int LinearIncrementor::valueAt(int index) const
{
return m_values[index];
}
void LinearIncrementor::incrementAt(int index)
{
const int n = static_cast<int>(m_values.size());
const int value = m_values[index];
while (index+1 < n && value == m_values[index+1])
++index;
++m_values[index];
}
RunList::RunList() : m_firstFree(-1)
{
}
int RunList::allocateRun(int endIndex, int value)
{
int runIndex = -1;
if (m_firstFree == -1)
{
runIndex = static_cast<int>(m_runs.size());
m_runs.resize(runIndex + 1);
}
else
{
runIndex = m_firstFree;
m_firstFree = m_runs[runIndex].m_end;
}
Run& run = m_runs[runIndex];
run.m_end = endIndex;
run.m_value = value;
return runIndex;
}
void RunList::freeRun(int index)
{
m_runs[index].m_end = m_firstFree;
m_firstFree = index;
}
RunList::Run& RunList::runAt(int index)
{
return m_runs[index];
}
const RunList::Run& RunList::runAt(int index) const
{
return m_runs[index];
}
ConstantIncrementor::ConstantIncrementor(int numElems) : m_runIndices(numElems, 0)
{
const int runIndex = m_runs.allocateRun(numElems-1, 0);
assert(runIndex == 0);
}
int ConstantIncrementor::valueAt(int index) const
{
return m_runs.runAt(m_runIndices[index]).m_value;
}
void ConstantIncrementor::incrementAt(int index)
{
const int numElems = static_cast<int>(m_runIndices.size());
const int curRunIndex = m_runIndices[index];
RunList::Run& curRun = m_runs.runAt(curRunIndex);
index = curRun.m_end;
const bool freeCurRun = index == 0 || m_runIndices[index-1] != curRunIndex;
RunList::Run* runToMerge = NULL;
int runToMergeIndex = -1;
if (curRun.m_end+1 < numElems)
{
const int nextRunIndex = m_runIndices[curRun.m_end+1];
RunList::Run& nextRun = m_runs.runAt(nextRunIndex);
if (curRun.m_value+1 == nextRun.m_value)
{
runToMerge = &nextRun;
runToMergeIndex = nextRunIndex;
}
}
if (freeCurRun && !runToMerge) // then free and allocate at the same time
{
++curRun.m_value;
}
else
{
if (freeCurRun)
{
m_runs.freeRun(curRunIndex);
}
else
{
--curRun.m_end;
}
if (runToMerge)
{
m_runIndices[index] = runToMergeIndex;
}
else
{
m_runIndices[index] = m_runs.allocateRun(index, curRun.m_value+1);
}
}
}
int main(int argc, char* argv[])
{
const int numElems = 100;
const int numInc = 1000000;
LinearIncrementor linearInc(numElems);
ConstantIncrementor constInc(numElems);
srand(1);
for (int i = 0; i < numInc; ++i)
{
const int index = rand() % numElems;
linearInc.incrementAt(index);
constInc.incrementAt(index);
for (int j = 0; j < numElems; ++j)
{
if (linearInc.valueAt(j) != constInc.valueAt(j))
{
printf("Error: differing values at increment step %d, value at index %d\n", i, j);
}
}
}
return 0;
}
As a complement to the other answers: if you can only have the array, then you cannot indeed guarantee the operation will be constant-time; but because the array is sorted, you can find the end of a run of identical numbers in log n operations, not in n operations. This is simply a binary search.
If we expect most runs of numbers to be short, we should use galloping search, which is a variant where we first find the bounds by looking at positions +1, +2, +4, +8, +16, etc. and then doing binary search inside. You would get a time that is often constant (and extremely fast if the item is unique) but can grow up to log n. Unless for some reason long runs of identical numbers remain common even after many updates, this might outperform any solution that requires keeping additional data.

Using an array and moving duplicates to end

I got this question at an interview and at the end was told there was a more efficient way to do this but have still not been able to figure it out. You are passing into a function an array of integers and an integer for size of array. In the array you have a lot of numbers, some that repeat for example 1,7,4,8,2,6,8,3,7,9,10. You want to take that array and return an array where all the repeated numbers are put at the end of the array so the above array would turn into 1,7,4,8,2,6,3,9,10,8,7. The numbers I used are not important and I could not use a buffer array. I was going to use a BST, but the order of the numbers must be maintained(except for the duplicate numbers). I could not figure out how to use a hash table so I ended up using a double for loop(n^2 horrible I know). How would I do this more efficiently using c++. Not looking for code, just an idea of how to do it better.
In what follows:
arr is the input array;
seen is a hash set of numbers already encountered;
l is the index where the next unique element will be placed;
r is the index of the next element to be considered.
Since you're not looking for code, here is a pseudo-code solution (which happens to be valid Python):
arr = [1,7,4,8,2,6,8,3,7,9,10]
seen = set()
l = 0
r = 0
while True:
# advance `r` to the next not-yet-seen number
while r < len(arr) and arr[r] in seen:
r += 1
if r == len(arr): break
# add the number to the set
seen.add(arr[r])
# swap arr[l] with arr[r]
arr[l], arr[r] = arr[r], arr[l]
# advance `l`
l += 1
print arr
On your test case, this produces
[1, 7, 4, 8, 2, 6, 3, 9, 10, 8, 7]
I would use an additional map, where the key is the integer value from the array and the value is an integer set to 0 in the beginning. Now I would go through the array and increase the values in the map if the key is already in the map.
In the end I would go again through the array. When the integer from the array has a value of one in the map, I would not change anything. When it has a value of 2 or more in the map I would swap the integer from the array with the last one.
This should result in a runtime of O(n*log(n))
The way I would do this would be to create an array twice the size of the original and create a set of integers.
Then Loop through the original array, add each element to the set, if it already exists add it to the 2nd half of the new array, else add it to the first half of the new array.
In the end you would get an array that looks like: (using your example)
1,7,4,8,2,6,3,9,10,-,-,8,7,-,-,-,-,-,-,-,-,-
Then I would loop through the original array again and make each spot equal to the next non-null position (or 0'd or whatever you decided)
That would make the original array turn into your solution...
This ends up being O(n) which is about as efficient as I can think of
Edit: since you can not use another array, when you find a value that is already in the
set you can move every value after it forward one and set the last value equal to the
number you just checked, this would in effect do the same thing but with a lot more operations.
I have been out of touch for a while, but I'd probably start out with something like this and see how it scales with larger input. I know you didn't ask for code but in some cases it's easier to understand than an explanation.
Edit: Sorry I missed the requirement that you cannot use a buffer array.
// returns new vector with dupes a the end
std::vector<int> move_dupes_to_end(std::vector<int> input)
{
std::set<int> counter;
std::vector<int> result;
std::vector<int> repeats;
for (std::vector<int>::iterator i = input.begin(); i < input.end(); i++)
{
if (counter.find(*i) == counter.end())
result.push_back(*i);
else
repeats.push_back(*i);
counter.insert(*i);
}
result.insert(result.end(), repeats.begin(), repeats.end());
return result;
}
#include <algorithm>
T * array = [your array];
size_t size = [array size];
// Complexity:
sort( array, array + size ); // n * log(n) and could be threaded
// (if merge sort)
T * last = unique( array, array + size ); // n, but the elements after the last
// unique element are not defined
Check sort and unique.
void remove_dup(int* data, int count) {
int* L=data; //place to put next unique number
int* R=data+count; //place to place next repeat number
std::unordered_set<int> found(count); //keep track of what's been seen
for(int* cur=data; cur<R; ++cur) { //until we reach repeats
if(found.insert(*cur).second == false) { //if we've seen it
std::swap(*cur,*--R); //put at the beginning of the repeats
} else //or else
std::swap(*cur,*L++); //put it next in the unique list
}
std::reverse(R, data+count); //reverse the repeats to be in origional order
}
http://ideone.com/3choA
Not that I would turn in code this poorly commented. Also note that unordered_set probably uses it's own array internally, bigger than data. (This has been rewritten based on aix's answer, to be much faster)
If you know the bounds on what the integer values are, B, and the size of the integer array, SZ, then you can do something like the following:
Create an array of booleans seen_before with B elements, initialized to 0.
Create a result array result of integers with SZ elements.
Create two integers, one for front_pos = 0, one for back_pos = SZ - 1.
Iterate across the original list:
Set an integer variable val to the value of the current element
If seen_before[val] is set to 1, put the number at result[back_pos] then decrement back_pos
If seen_before[val] is not set to 1, put the number at result[front_pos] then increment front_pos and set seen_before[val] to 1.
Once you finish iterating across the main list, all the unique numbers will be at the front of the list while the duplicate numbers will be at the back. Fun part is that the entire process is done in one pass. Note that this only works if you know the bounds of the values appearing in the original array.
Edit: It was pointed out that there's no bounds on the integers used, so instead of initializing seen_before as an array with B elements, initialize it as a map<int, bool>, then continue as usual. That should get you n*log(n) performance.
This can be done by iterating the array & marking index of the first change.
later on swaping that mark index value with next unique value
& then incrementing that mark index for next swap
Java Implementation:
public static void solve() {
Integer[] arr = new Integer[] { 1, 7, 4, 8, 2, 6, 8, 3, 7, 9, 10 };
final HashSet<Integer> seen = new HashSet<Integer>();
int l = -1;
for (int i = 0; i < arr.length; i++) {
if (seen.contains(arr[i])) {
if (l == -1) {
l = i;
}
continue;
}
if (l > -1) {
final int temp = arr[i];
arr[i] = arr[l];
arr[l] = temp;
l++;
}
seen.add(arr[i]);
}
}
output is 1 7 4 8 2 6 3 9 10 8 7
It's ugly, but it meets the requirements of moving the duplicates to the end in place (no buffer array)
// warning, some light C++11
void dup2end(int* arr, size_t cnt)
{
std::set<int> k;
auto end = arr + cnt-1;
auto max = arr + cnt;
auto curr = arr;
while(curr < max)
{
auto res = k.insert(*curr);
// first time encountered
if(res.second)
{
++curr;
}
else
{
// duplicate:
std::swap(*curr, *end);
--end;
--max;
}
}
}
void move_duplicates_to_end(vector<int> &A) {
if(A.empty()) return;
int i = 0, tail = A.size()-1;
while(i <= tail) {
bool is_first = true; // check of current number is first-shown
for(int k=0; k<i; k++) { // always compare with numbers before A[i]
if(A[k] == A[i]) {
is_first = false;
break;
}
}
if(is_first == true) i++;
else {
int tmp = A[i]; // swap with tail
A[i] = A[tail];
A[tail] = tmp;
tail--;
}
}
If the input array is {1,7,4,8,2,6,8,3,7,9,10}, then the output is {1,7,4,8,2,6,10,3,9,7,8}. Comparing with your answer {1,7,4,8,2,6,3,9,10,8,7}, the first half is the same, while the right half is different, because I swap all duplicates with the tail of the array. As you mentioned, the order of the duplicates can be arbitrary.