|
| 1 | +--- |
| 2 | +id: ac-grouping |
| 3 | +source: AtCoder |
| 4 | +title: Grouping |
| 5 | +author: Kiruthees G |
| 6 | +--- |
| 7 | + |
| 8 | +[Unofficial Analysis (C++)](https://nwatx.me/post/atcoderdp) |
| 9 | + |
| 10 | +## Explanation |
| 11 | + |
| 12 | +We define $dp[i]$ as the maximum score possible for the subset of rabbits represented by mask $i$. |
| 13 | + |
| 14 | +The transition involves two steps for each mask $i$: |
| 15 | + |
| 16 | +1. **Base Score**: Initialize $dp[i]$ by calculating the sum of $a_{j,k}$ for all pairs $(j, k)$ within the mask. This assumes all rabbits in the mask form a single group. |
| 17 | +2. **Merging Subsets**: Improve $dp[i]$ by splitting the mask into two disjoint submasks $j$ and $i \oplus j$. The transition is: |
| 18 | + |
| 19 | +$$ |
| 20 | +dp[i] = \max_{j \subseteq i} \left(dp[i], dp[j] + dp[i \oplus j] \right) |
| 21 | +$$ |
| 22 | + |
| 23 | + |
| 24 | +<Info> |
| 25 | + |
| 26 | +Since $j$ is a subset of $i$, the XOR operation ($i \oplus j$) acts like set subtraction. It "toggles off" the bits that are present in $j$, leaving you with exactly the remaining bits of $i$. This guarantees that $j$ and $i \oplus j$ are disjoint parts that perfectly combine to form $i$. |
| 27 | + |
| 28 | +For an efficient way to iterate over submasks $j$, we use the pattern `for (int j = i; j; j = (j - 1) & i)`. You can read more about this technique in the [Merging Subsets section](/gold/dp-bitmasks#merging-subsets) of the module. |
| 29 | + |
| 30 | +</Info> |
| 31 | + |
| 32 | + |
| 33 | +## Implementation |
| 34 | + |
| 35 | +**Time Complexity**: $\mathcal{O}(3^N + N^2 \cdot 2^N)$ |
| 36 | + |
| 37 | +<LanguageSection> |
| 38 | +<CPPSection> |
| 39 | + |
| 40 | +```cpp |
| 41 | +#include <bits/stdc++.h> |
| 42 | +using namespace std; |
| 43 | + |
| 44 | +using ll = long long; |
| 45 | + |
| 46 | +int main() { |
| 47 | + ios_base::sync_with_stdio(false); |
| 48 | + cin.tie(nullptr); |
| 49 | + |
| 50 | + int n; |
| 51 | + cin >> n; |
| 52 | + vector a(n, vector<int>(n)); |
| 53 | + for (int i = 0; i < n; i++) { |
| 54 | + for (int j = 0; j < n; j++) { cin >> a[i][j]; } |
| 55 | + } |
| 56 | + |
| 57 | + vector<ll> dp(1 << n); |
| 58 | + |
| 59 | + // Iterate through every possible subset of rabbits |
| 60 | + for (int i = 1; i < (1 << n); i++) { |
| 61 | + // Calculate internal score if this subset forms ONE single group |
| 62 | + for (int j = 0; j < n; j++) { |
| 63 | + if (!(i >> j & 1)) continue; |
| 64 | + for (int k = j + 1; k < n; k++) { |
| 65 | + if (!(i >> k & 1)) continue; |
| 66 | + dp[i] += a[j][k]; |
| 67 | + } |
| 68 | + } |
| 69 | + |
| 70 | + // Iterate over all submasks j of the current mask i |
| 71 | + // Try splitting the mask into smaller groups to find a better score |
| 72 | + for (int j = (i - 1) & i; j > 0; j = (j - 1) & i) { |
| 73 | + dp[i] = max(dp[i], dp[j] + dp[i ^ j]); |
| 74 | + } |
| 75 | + } |
| 76 | + |
| 77 | + cout << dp[(1 << n) - 1] << '\n'; |
| 78 | +} |
| 79 | +``` |
| 80 | +
|
| 81 | +</CPPSection> |
| 82 | +</LanguageSection> |
0 commit comments