code_fille's blog

Find count of subarrays having sum in a given range in less than O(n^2).
By code_fille, history, 9 years ago, In English

Given an array of integers and a range find the count of subarrays whose sum lies in the given range. Do so in less than O(n^2) complexity.

  • Vote: I like it
  • +5
  • Vote: I do not like it

Comments Comments (17)
| Write comment?
»
9 years ago, # |
← Rev. 6   Vote: I like it +10 Vote: I do not like it

let us call the range [ l, r ] build the prefix sum array for the given array, let us call it B (B[0]=0 for empty prefix)

now
B[i] - B[j] where 0 ≤ j < i
gives all the sums of subarrays that end in ith element

So, in order for the sum of subarray [ j + 1...i ] to be between l and r the following should be true:

l ≤ B[i] - B[j] ≤ r
=>
l - B[i] ≤  - B[j] ≤ r - B[i]
=>
B[i] - l ≥ B[j] ≥ B[i] - r

so for each B[i], we should find the number of B[j] s that are in the previous range and j < i, which is done in a segment tree

total complexity: O(nlogn)

NOTE If the values are all positive you can use binary search for O(nlogn)

→ Reply
  • »
    »
    9 years ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    How could segment tree be used to find the possible values of j?

    → Reply
    • »
      »
      »
      9 years ago, # ^ |
      ← Rev. 3   Vote: I like it 0 Vote: I do not like it

      Array C , C[i] is the number of B[j]'s which are equal to i
      You need a segment tree to perform 2 types of queries:
      1) add 1 to C[x]
      2) get the sum of C[x] : s<=x<=e

      So for each B[i] you query for the sum of C[x] between B[i]-r and B[i]-l Then you add 1 to C[B[i]]

      → Reply
      • »
        »
        »
        »
        7 years ago, # ^ |
          Vote: I like it 0 Vote: I do not like it

        B[i] can be negative ..how it is possible to store in array?

        → Reply
        • »
          »
          »
          »
          »
          7 years ago, # ^ |
            Vote: I like it 0 Vote: I do not like it

          Then compress the values

          → Reply
          • »
            »
            »
            »
            »
            »
            7 years ago, # ^ |
              Vote: I like it 0 Vote: I do not like it

            I was trying to solve this question. I tried to compress the values. But it is giving WA. Here is the code. Please help me.

            → Reply
            • »
              »
              »
              »
              »
              »
              »
              7 years ago, # ^ |
                Vote: I like it 0 Vote: I do not like it

              You will have at most N+2 different values (Including L and R), compress them.

              → Reply
              • »
                »
                »
                »
                »
                »
                »
                »
                7 years ago, # ^ |
                  Vote: I like it 0 Vote: I do not like it

                Can you please have a glance at my code?

                → Reply
              • »
                »
                »
                »
                »
                »
                »
                »
                7 years ago, # ^ |
                  Vote: I like it 0 Vote: I do not like it

                For each B[i], I have to put B[i]-l and B[i]-r. I have to compress n*2*n values.

                → Reply
                • »
                  »
                  »
                  »
                  »
                  »
                  »
                  »
                  »
                  7 years ago, # ^ |
                    Vote: I like it 0 Vote: I do not like it

                  Were you able to solve that question ? If yes, can you please share the code? I am also getting W.A.

                  → Reply
                • »
                  »
                  »
                  »
                  »
                  »
                  »
                  »
                  »
                  7 years ago, # ^ |
                    Vote: I like it 0 Vote: I do not like it

                  Yes, I solved it Here is the code

                  → Reply
  • »
    »
    9 years ago, # ^ |
    ← Rev. 2   Vote: I like it 0 Vote: I do not like it

    If the values are all positive you can solve in O(N) using only 3 pointers.

    Note: You can also solve in O(NlogN) without segment tree, just using divide-and-conquer idea.

    → Reply
    • »
      »
      »
      9 years ago, # ^ |
        Vote: I like it 0 Vote: I do not like it

      Could you elaborate how divide and conquer could be used?

      → Reply
      • »
        »
        »
        »
        9 years ago, # ^ |
        ← Rev. 3   Vote: I like it 0 Vote: I do not like it

        B — partial sum, B[0] = 0.

        Now we are going to split our verctor a into 2 parts with size N / 2 each.

        2 cases are possible:

        1) l, r < N / 2 or l, r > N / 2: We can find the answer recursively for the left part and for the right part.

        2) l < N / 2 and r > N / 2: We can find the answer in O(N) using 3 pointers but only if the left part sorted and the right part sorted too. But you can maintain these parts sorted by using merge sort algorithm.

        The algorithm is very similar to the algorithm for counting inversions in O(NlogN) time. You can find the explanation on coursera: link

        → Reply
  • »
    »
    8 years ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    For segment tree , Is the time complexity is (n*log(sum of all array elements)) ? correct me if i am wrong.

    → Reply
    • »
      »
      »
      8 years ago, # ^ |
        Vote: I like it 0 Vote: I do not like it

      correct me if i am wrong.

      Wrong? You didn't state anything.

      → Reply
  • »
    »
    6 years ago, # ^ |
      Vote: I like it +8 Vote: I do not like it

    Your solution is quite beautiful :)

    → Reply