How to determine the longest increasing subsequenc

Question

I have a set of integers. I want to find the longest increasing subsequence of that set using dynamic programming.

Answer 1

The following C++ implementation includes also some code that builds the actual longest increasing subsequence using an array called prev.

std::vector<int> longest_increasing_subsequence (const std::vector<int>& s)
{
    int best_end = 0;
    int sz = s.size();

    if (!sz)
        return std::vector<int>();

    std::vector<int> prev(sz,-1);
    std::vector<int> memo(sz, 0);

    int max_length = std::numeric_limits<int>::min();

    memo[0] = 1;

    for ( auto i = 1; i < sz; ++i)
    {
        for ( auto j = 0; j < i; ++j)
        {
            if ( s[j] < s[i] && memo[i] < memo[j] + 1 )
            {
                memo[i] =  memo[j] + 1;
                prev[i] =  j;
            }
        }

        if ( memo[i] > max_length ) 
        {
            best_end = i;
            max_length = memo[i];
        }
    }

    // Code that builds the longest increasing subsequence using "prev"
    std::vector<int> results;
    results.reserve(sz);

    std::stack<int> stk;
    int current = best_end;

    while (current != -1)
    {
        stk.push(s[current]);
        current = prev[current];
    }

    while (!stk.empty())
    {
        results.push_back(stk.top());
        stk.pop();
    }

    return results;
}

Implementation with no stack just reverse the vector

#include <iostream>
#include <vector>
#include <limits>
std::vector<int> LIS( const std::vector<int> &v ) {
  auto sz = v.size();
  if(!sz)
    return v;
  std::vector<int> memo(sz, 0);
  std::vector<int> prev(sz, -1);
  memo[0] = 1;
  int best_end = 0;
  int max_length = std::numeric_limits<int>::min();
  for (auto i = 1; i < sz; ++i) {
    for ( auto j = 0; j < i ; ++j) {
      if (s[j] < s[i] && memo[i] < memo[j] + 1) {
        memo[i] = memo[j] + 1;
        prev[i] = j;
      }
    }
    if(memo[i] > max_length) {
      best_end = i;
      max_length = memo[i];
    }
  }

  // create results
  std::vector<int> results;
  results.reserve(v.size());
  auto current = best_end;
  while (current != -1) {
    results.push_back(s[current]);
    current = prev[current];
  }
  std::reverse(results.begin(), results.end());
  return results;
}

Answer 2

Here are three steps of evaluating the problem from dynamic programming point of view:

1. Recurrence definition: maxLength(i) == 1 + maxLength(j) where 0 < j < i and array[i] > array[j]
2. Recurrence parameter boundary: there might be 0 to i - 1 sub-sequences passed as a paramter
3. Evaluation order: as it is increasing sub-sequence, it has to be evaluated from 0 to n

If we take as an example sequence {0, 8, 2, 3, 7, 9}, at index:

• [0] we'll get subsequence {0} as a base case
• [1] we have 1 new subsequence {0, 8}
• [2] trying to evaluate two new sequences {0, 8, 2} and {0, 2} by adding element at index 2 to existing sub-sequences - only one is valid, so adding third possible sequence {0, 2} only to parameters list ...

Here's the working C++11 code:

#include <iostream>
#include <vector>

int getLongestIncSub(const std::vector<int> &sequence, size_t index, std::vector<std::vector<int>> &sub) {
    if(index == 0) {
        sub.push_back(std::vector<int>{sequence[0]});
        return 1;
    }

    size_t longestSubSeq = getLongestIncSub(sequence, index - 1, sub);
    std::vector<std::vector<int>> tmpSubSeq;
    for(std::vector<int> &subSeq : sub) {
        if(subSeq[subSeq.size() - 1] < sequence[index]) {
            std::vector<int> newSeq(subSeq);
            newSeq.push_back(sequence[index]);
            longestSubSeq = std::max(longestSubSeq, newSeq.size());
            tmpSubSeq.push_back(newSeq);
        }
    }
    std::copy(tmpSubSeq.begin(), tmpSubSeq.end(),
              std::back_insert_iterator<std::vector<std::vector<int>>>(sub));

    return longestSubSeq;
}

int getLongestIncSub(const std::vector<int> &sequence) {
    std::vector<std::vector<int>> sub;
    return getLongestIncSub(sequence, sequence.size() - 1, sub);
}

int main()
{
    std::vector<int> seq{0, 8, 2, 3, 7, 9};
    std::cout << getLongestIncSub(seq);
    return 0;
}

Answer 3

def longestincrsub(arr1):
    n=len(arr1)
    l=[1]*n
    for i in range(0,n):
        for j in range(0,i)  :
            if arr1[j]<arr1[i] and l[i]<l[j] + 1:
                l[i] =l[j] + 1
    l.sort()
    return l[-1]
arr1=[10,22,9,33,21,50,41,60]
a=longestincrsub(arr1)
print(a)

even though there is a way by which you can solve this in O(nlogn) time(this solves in O(n^2) time) but still this way gives the dynamic programming approach which is also good .

Answer 4

OK, I will describe first the simplest solution which is O(N^2), where N is the size of the collection. There also exists a O(N log N) solution, which I will describe also. Look here for it at the section Efficient algorithms.

I will assume the indices of the array are from 0 to N - 1. So let's define DP[i] to be the length of the LIS (Longest increasing subsequence) which is ending at element with index i. To compute DP[i] we look at all indices j < i and check both if DP[j] + 1 > DP[i] and array[j] < array[i] (we want it to be increasing). If this is true we can update the current optimum for DP[i]. To find the global optimum for the array you can take the maximum value from DP[0...N - 1].

int maxLength = 1, bestEnd = 0;
DP[0] = 1;
prev[0] = -1;

for (int i = 1; i < N; i++)
{
   DP[i] = 1;
   prev[i] = -1;

   for (int j = i - 1; j >= 0; j--)
      if (DP[j] + 1 > DP[i] && array[j] < array[i])
      {
         DP[i] = DP[j] + 1;
         prev[i] = j;
      }

   if (DP[i] > maxLength)
   {
      bestEnd = i;
      maxLength = DP[i];
   }
}

I use the array prev to be able later to find the actual sequence not only its length. Just go back recursively from bestEnd in a loop using prev[bestEnd]. The -1 value is a sign to stop.

---

OK, now to the more efficient O(N log N) solution:

Let S[pos] be defined as the smallest integer that ends an increasing sequence of length pos. Now iterate through every integer X of the input set and do the following:

1. If X > last element in S, then append X to the end of S. This essentialy means we have found a new largest LIS.

2. Otherwise find the smallest element in S, which is >= than X, and change it to X. Because S is sorted at any time, the element can be found using binary search in log(N).

Total runtime - N integers and a binary search for each of them - N * log(N) = O(N log N)

Now let's do a real example:

Collection of integers: 2 6 3 4 1 2 9 5 8

Steps:

0. S = {} - Initialize S to the empty set
1. S = {2} - New largest LIS
2. S = {2, 6} - New largest LIS
3. S = {2, 3} - Changed 6 to 3
4. S = {2, 3, 4} - New largest LIS
5. S = {1, 3, 4} - Changed 2 to 1
6. S = {1, 2, 4} - Changed 3 to 2
7. S = {1, 2, 4, 9} - New largest LIS
8. S = {1, 2, 4, 5} - Changed 9 to 5
9. S = {1, 2, 4, 5, 8} - New largest LIS

So the length of the LIS is 5 (the size of S).

To reconstruct the actual LIS we will again use a parent array. Let parent[i] be the predecessor of element with index i in the LIS ending at element with index i.

To make things simpler, we can keep in the array S, not the actual integers, but their indices(positions) in the set. We do not keep {1, 2, 4, 5, 8}, but keep {4, 5, 3, 7, 8}.

That is input[4] = 1, input[5] = 2, input[3] = 4, input[7] = 5, input[8] = 8.

If we update properly the parent array, the actual LIS is:

input[S[lastElementOfS]], 
input[parent[S[lastElementOfS]]],
input[parent[parent[S[lastElementOfS]]]],
........................................

Now to the important thing - how do we update the parent array? There are two options:

1. If X > last element in S, then parent[indexX] = indexLastElement. This means the parent of the newest element is the last element. We just prepend X to the end of S.

2. Otherwise find the index of the smallest element in S, which is >= than X, and change it to X. Here parent[indexX] = S[index - 1].

Answer 5

This can be solved in O(n^2) using dynamic programming.

Process the input elements in order and maintain a list of tuples for each element. Each tuple (A,B), for the element i will denotes, A = length of longest increasing sub-sequence ending at i and B = index of predecessor of list[i] in the longest increasing sub-sequence ending at list[i].

Start from element 1, the list of tuple for element 1 will be [(1,0)] for element i, scan the list 0..i and find element list[k] such that list[k] < list[i], the value of A for element i, Ai will be Ak + 1 and Bi will be k. If there are multiple such elements, add them to the list of tuples for element i.

In the end, find all the elements with max value of A (length of LIS ending at element) and backtrack using the tuples to get the list.

I have shared the code for same at http://www.edufyme.com/code/?id=66f041e16a60928b05a7e228a89c3799

Answer 6

Speaking about DP solution, I found it surprising that no one mentioned the fact that LIS can be reduced to LCS. All you need to do is sort the copy of the original sequence, remove all the duplicates and do LCS of them. In pseudocode it is:

def LIS(S):
    T = sort(S)
    T = removeDuplicates(T)
    return LCS(S, T)

And the full implementation written in Go. You do not need to maintain the whole n^2 DP matrix if you do not need to reconstruct the solution.

func lcs(arr1 []int) int {
    arr2 := make([]int, len(arr1))
    for i, v := range arr1 {
        arr2[i] = v
    }
    sort.Ints(arr1)
    arr3 := []int{}
    prev := arr1[0] - 1
    for _, v := range arr1 {
        if v != prev {
            prev = v
            arr3 = append(arr3, v)
        }
    }

    n1, n2 := len(arr1), len(arr3)

    M := make([][]int, n2 + 1)
    e := make([]int, (n1 + 1) * (n2 + 1))
    for i := range M {
        M[i] = e[i * (n1 + 1):(i + 1) * (n1 + 1)]
    }

    for i := 1; i <= n2; i++ {
        for j := 1; j <= n1; j++ {
            if arr2[j - 1] == arr3[i - 1] {
                M[i][j] = M[i - 1][j - 1] + 1
            } else if M[i - 1][j] > M[i][j - 1] {
                M[i][j] = M[i - 1][j]
            } else {
                M[i][j] = M[i][j - 1]
            }
        }
    }

    return M[n2][n1]
}