This page looks best with JavaScript enabled

Knapsack Problems

 ·   ·  โ˜• 3 min read  ·  ๐Ÿฆ‚ Kyle · ๐Ÿ‘€... views

knapsack problem and its variations

0-1 knapsack problem

description

Given a set of $n$ items, each with a weight $w_{i}$ and a value $v_{i}$, along with a maximum weight capacity $W$, try pick some of these items (total weight not surpassing $W$), so that the total value is maximized, find the maximum total value.

solution to acwing: 01่ƒŒๅŒ…้—ฎ้ข˜

problem link
time: $O(NV)$, space: $O(V)$
for every item, total weight of items in our knapsack decrease (so that we take each item at most once) from maximum (knapsack volume) to current item’s weight.
dp[v] is the maximum value we get with total weight of $v$ (or less) if we do not take in this item;
dp[v-p.first] + p.second is the maximum value we get with total weight of $v$ (or less) if we take this item.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
#include <bits/stdc++.h>

using namespace std;

static int x = []() { std::ios::sync_with_stdio(false); cin.tie(0); return 0; }();
typedef long long ll;

int N, V;

int solve(vector<pair<int, int>> &A) {
  vector<int> dp(V + 1);
  for (auto &p : A) {
    for (int v = V; v >= p.first; --v) {
      dp[v] = max(dp[v], dp[v - p.first] + p.second);
    }
  }
  return *max_element(dp.begin(), dp.end());
}

int main(int argc, char const *argv[]) {
  cin >> N >> V;  // volume
  vector<pair<int, int>> A(N);
  for (int i = 0; i < N; ++i) {
    cin >> A[i].first >> A[i].second;  // (weight, value)... not (value, weight)
  }
  cout << solve(A) << endl;
  return 0;
}

count ways / check valid way exists

leetcode: 416 Partition Equal Subset Sum
time: $O(NV)$, space: $O(V)$
now we need to count the number of ways to form exact $V$ total weight
similar as above, but initialize dp[0]=1, and rest as 0.
if we do not need to count the number of ways, but just check if there exists one valid way, use or operation indead of +.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
// cannot count ways, gets overflow, use '|' operation is enough
class Solution {
 public:
  bool canPartition(vector<int>& nums) {
    int sum = accumulate(nums.begin(), nums.end(), 0);
    if (sum & 1) return false;
    vector<int> dp(sum / 2 + 1);
    dp[0] = 1;
    for (int num : nums) {
      for (int i = sum / 2; i >= num; --i) {
        dp[i] |= dp[i - num];
        // dp[i] += dp[i - num]; // overflow, if number of ways is too big
      }
    }
    return dp[sum / 2];
  }
};
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
// for 0-1 knapsack problem, using bitset is more simple -- update all bits same time
// use bitset
// https://leetcode.com/problems/partition-equal-subset-sum/discuss/90590/Simple-C%2B%2B-4-line-solution-using-a-bitset
class Solution {
 public:
  bool canPartition(vector<int>& nums) {
    bitset<10001> bits(1);
    int sum = accumulate(nums.begin(), nums.end(), 0);
    for (int num : nums) bits |= bits << num;
    return !(sum & 1) && bits[sum >> 1];
  }
};

get a valid path

Often in 0-1 knapsack problems, we want to know exactly the what set of objects we take (maybe multiple, but we just need any one of them).

To achieve this, we only need an extra array of size V (which is the volume of our knapsack). Idea is that we store at path[i] the object that first let us arrive volume i.

(haven’t found a proper/simple example problem yet… to be added)

If we want to get a valid set of integers (objects) in the above problem: leetcode: 416 Partition Equal Subset Sum

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
class Solution {
 public:
  bool canPartition(vector<int>& nums) {
    bitset<10001> bits(1);
    int sum = accumulate(nums.begin(), nums.end(), 0);
    vector<int> path((sum >> 1) + 1);
    for (int num : nums) {
      auto change = bits;
      bits |= bits << num;
      change ^= bits; // set bits that changed in this loop
      for (int i = change._Find_first(); i <= a; i = change._Find_next(i) {
        path[i] = num;
      }
    }
    if (!(sum & 1) && bits[sum >> 1]) {
      for (int j = sum >> 1; j; ) {
        cout << path[j] << " ";
        j -= path[j];
      }
      cout << endl;
    }
    return !(sum & 1) && bits[sum >> 1];
  }
};

unbounded knapsack problem

description

similar to the 0-1 knapsack problem, but now there are infinite number of each item, so we can take multiple of each.

solution to acwing: ๅฎŒๅ…จ่ƒŒๅŒ…้—ฎ้ข˜

problem link
time: $O(NV)$, space: $O(V)$
only difference from the 0-1 knapsack problem solution is that now for each item, total weight of items in our knapsack increases (so we can re-put this item in our knapsack) from current item’s weight to maximum (knapsack’s volume)

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
#include <bits/stdc++.h>

using namespace std;

static int x = []() { std::ios::sync_with_stdio(false); cin.tie(0); return 0; }();
typedef long long ll;

int N, V;

int solve(vector<pair<int, int>> &A) {
  vector<int> dp(V + 1);
  for (auto &p : A) {
    for (int v = p.first; v <= V; ++v) { // only difference from 0-1 knapsack problem
      dp[v] = max(dp[v], dp[v - p.first] + p.second);
    }
  }
  return dp[V];
}

int main(int argc, char const *argv[]) {
  cin >> N >> V;  // volume
  vector<pair<int, int>> A(N);
  for (int i = 0; i < N; ++i) {
    cin >> A[i].first >> A[i].second;  // (weight, value)... not (value, weight)
  }
  cout << solve(A) << endl;
  return 0;
}

count ways / check valid way exists

leetcode: 518 Coin Change 2
similar to the previous one, but we update from weight $w_i$ to $V$

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
class Solution {
 public:
  int change(int t, vector<int>& cs) {
    // int dp[t + 1] = {1};
    vector<int> dp(t + 1);
    dp[0] = 1;
    for (auto c : cs)
      for (auto j = c; j <= t; ++j) dp[j] += dp[j - c];
    return dp[t];
  }
};

bounded knapsack problem

description

n types of goods, there are $s_i$ number of type $i$ goods and each with weight $w_i$ and value $v_i$.

solution to acwing: ๅคš้‡่ƒŒๅŒ…้—ฎ้ข˜ I

problem link
time: $O(NVlogM)$, space: $O(V)$
reduce bounded knapsack problem to 0-1 knapsack problem
for type $i$ goods (M – total number), split them into groups ($ceil(log_2M)$ groups) of number $1, 2, 4, 8, … 2^k, M-2^k$. and treat these groups as 0-1 knapsack problem items.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
#include <bits/stdc++.h>

using namespace std;

static int x = []() { std::ios::sync_with_stdio(false); cin.tie(0); return 0; }();
typedef long long ll;

int N, V;

int solve(vector<int> &weights, vector<int> &values, vector<int> &nums) {
  vector<int> dp(V + 1);
  for (int i = 0; i < N; ++i) {
    int num = min(nums[i], V / weights[i]);
    for (int k = 1; num; k *= 2) {
      if (k > num) k = num;
      num -= k;
      for (int j = V; j >= weights[i] * k; --j) {
        dp[j] = max(dp[j], dp[j - weights[i] * k] + values[i] * k);
      }
    }
  }
  return dp[V];
}

int main(int argc, char const *argv[]) {
  cin >> N >> V;  // volume
  vector<int> weights(N), values(N), nums(N);
  for (int i = 0; i < N; ++i) {
    cin >> weights[i] >> values[i] >>
        nums[i];  // (weight, value)... not (value, weight)
  }
  cout << solve(weights, values, nums) << endl;
  return 0;
}

more knapsack problems for exercise

refs

Share on