#320545 (C++17) No.519 アイドルユニット

• 問題: No.519 アイドルユニット
• ユーザー: ei1333333
• 提出日時: 2019-03-02 23:17:06
• 言語: C++17 (gcc 12.3.0 + boost 1.83.0)
• 結果: AC
• 実行時間: 2 ms / 1,000 ms
• コード長: 24,876 bytes
• コンパイル時間: 3,845 ms
• コンパイル使用メモリ: 252,212 KB
• 実行使用メモリ: 6,944 KB
• 最終ジャッジ日時: 2024-06-23 13:04:30
• 合計ジャッジ時間: 4,891 ms
• ジャッジサーバーID （参考情報）: judge5 / judge1

入力	結果	実行時間; 実行使用メモリ
testcase_00	AC	2 ms; 6,816 KB
testcase_01	AC	2 ms; 6,940 KB
testcase_02	AC	2 ms; 6,940 KB
testcase_03	AC	2 ms; 6,944 KB
testcase_04	AC	2 ms; 6,944 KB
testcase_05	AC	2 ms; 6,940 KB
testcase_06	AC	2 ms; 6,940 KB
testcase_07	AC	2 ms; 6,940 KB
testcase_08	AC	2 ms; 6,940 KB
testcase_09	AC	2 ms; 6,940 KB
testcase_10	AC	2 ms; 6,940 KB
testcase_11	AC	2 ms; 6,944 KB
testcase_12	AC	2 ms; 6,944 KB
testcase_13	AC	2 ms; 6,944 KB
testcase_14	AC	2 ms; 6,944 KB
testcase_15	AC	2 ms; 6,944 KB
testcase_16	AC	2 ms; 6,940 KB
testcase_17	AC	2 ms; 6,940 KB
testcase_18	AC	2 ms; 6,944 KB
testcase_19	AC	2 ms; 6,940 KB
testcase_20	AC	2 ms; 6,940 KB
testcase_21	AC	2 ms; 6,944 KB
testcase_22	AC	2 ms; 6,944 KB
testcase_23	AC	2 ms; 6,944 KB
testcase_24	AC	2 ms; 6,940 KB
testcase_25	AC	2 ms; 6,944 KB
testcase_26	AC	2 ms; 6,944 KB
testcase_27	AC	2 ms; 6,940 KB
testcase_28	AC	2 ms; 6,940 KB
testcase_29	AC	2 ms; 6,944 KB
testcase_30	AC	2 ms; 6,940 KB
testcase_31	AC	2 ms; 6,940 KB
testcase_32	AC	2 ms; 6,944 KB
testcase_33	AC	2 ms; 6,940 KB

ソースコード

#include<bits/stdc++.h>

using namespace std;

// https://min-25.hatenablog.com/entry/2016/11/21/222625

template< typename CostType, typename TotalCostType=int64_t >
class MaximumWeightedMatching {
  /*
    Maximum Weighted Matching in General Graphs.
    - O(nm \log(n)) time
    - O(n + m) space
    Note: each vertex is 1-indexed.
    Ref:
      Harold N. Gabow,
      "Data Structures for Weighted Matching and
       Extensions to b-matching and f-factors" (2016)
      (https://arxiv.org/abs/1611.07541)
  */
public:
  using cost_t = CostType;
  using tcost_t = TotalCostType;

private:
  enum Label {
    kSeparated = -2, kInner = -1, kFree = 0, kOuter = 1
  };
  static constexpr cost_t Inf = cost_t(1) << (sizeof(cost_t) * 8 - 2);

private:
  template< typename T >
  class BinaryHeap {
  public:
    struct Node {
      bool operator<(const Node &rhs) const { return value < rhs.value; }

      T value;
      int id;
    };

    BinaryHeap() {}

    BinaryHeap(int N) : size_(0), node(N + 1), index(N, 0) {}

    int size() const { return size_; }

    bool empty() const { return size_ == 0; }

    void clear() { while(size_ > 0) index[node[size_--].id] = 0; }

    T min() const { return node[1].value; }

    int argmin() const { return node[1].id; } // argmin ?
    T get_val(int id) const { return node[index[id]].value; }

    void pop() { if(size_ > 0) pop(1); }

    void erase(int id) { if(index[id]) pop(index[id]); }

    bool has(int id) const { return index[id] != 0; }

    void update(int id, T v) {
      if(!has(id)) return push(id, v);
      bool up = (v < node[index[id]].value);
      node[index[id]].value = v;
      if(up) up_heap(index[id]);
      else down_heap(index[id]);
    }

    void decrease_key(int id, T v) {
      if(!has(id)) return push(id, v);
      if(v < node[index[id]].value) node[index[id]].value = v, up_heap(index[id]);
    }

    void push(int id, T v) {
      // assert(!has(id));
      index[id] = ++size_;
      node[size_] = {v, id};
      up_heap(size_);
    }

  private:
    void pop(int pos) {
      index[node[pos].id] = 0;
      if(pos == size_) {
        --size_;
        return;
      }
      bool up = (node[size_].value < node[pos].value);
      node[pos] = node[size_--];
      index[node[pos].id] = pos;
      if(up) up_heap(pos);
      else down_heap(pos);
    }

    void swap_node(int a, int b) {
      swap(node[a], node[b]);
      index[node[a].id] = a;
      index[node[b].id] = b;
    }

    void down_heap(int pos) {
      for(int k = pos, nk = k; 2 * k <= size_; k = nk) {
        if(node[2 * k] < node[nk]) nk = 2 * k;
        if(2 * k + 1 <= size_ && node[2 * k + 1] < node[nk]) nk = 2 * k + 1;
        if(nk == k) break;
        swap_node(k, nk);
      }
    }

    void up_heap(int pos) {
      for(int k = pos; k > 1 && node[k] < node[k >> 1]; k >>= 1) swap_node(k, k >> 1);
    }

    int size_;
    vector< Node > node;
    vector< int > index;
  };

  template< typename Key >
  class PairingHeaps {
  private:
    struct Node {
      Node() : prev(-1) {} // "prev < 0" means the node is unused.
      Node(Key v) : key(v), child(0), next(0), prev(0) {}

      Key key;
      int child, next, prev;
    };

  public:
    PairingHeaps(int H, int N) : heap(H), node(N) {
      // It consists of `H` Pairing heaps.
      // Each heap-node ID can appear at most 1 time(s) among heaps
      // and should be in [1, N).
    }

    void clear(int h) { if(heap[h]) clear_rec(heap[h]), heap[h] = 0; }

    void clear_all() {
      for(size_t i = 0; i < heap.size(); ++i) heap[i] = 0;
      for(size_t i = 0; i < node.size(); ++i) node[i] = Node();
    }

    bool empty(int h) const { return !heap[h]; }

    bool used(int v) const { return node[v].prev >= 0; }

    Key min(int h) const { return node[heap[h]].key; }

    int argmin(int h) const { return heap[h]; }

    void pop(int h) {
      // assert(!empty(h));
      erase(h, heap[h]);
    }

    void push(int h, int v, Key key) {
      // assert(!used(v));
      node[v] = Node(key);
      heap[h] = merge(heap[h], v);
    }

    void erase(int h, int v) {
      if(!used(v)) return;
      int w = two_pass_pairing(node[v].child);
      if(!node[v].prev) heap[h] = w;
      else {
        cut(v);
        heap[h] = merge(heap[h], w);
      }
      node[v].prev = -1;
    }

    void decrease_key(int h, int v, Key key) {
      if(!used(v)) return push(h, v, key);
      if(!node[v].prev) node[v].key = key;
      else {
        cut(v);
        node[v].key = key;
        heap[h] = merge(heap[h], v);
      }
    }

  private:
    void clear_rec(int v) {
      for(; v; v = node[v].next) {
        if(node[v].child) clear_rec(node[v].child);
        node[v].prev = -1;
      }
    }

    inline void cut(int v) {
      auto &n = node[v];
      int pv = n.prev, nv = n.next;
      auto &pn = node[pv];
      if(pn.child == v) pn.child = nv;
      else pn.next = nv;
      node[nv].prev = pv;
      n.next = n.prev = 0;
    }

    int merge(int l, int r) {
      if(!l) return r;
      if(!r) return l;
      if(node[l].key > node[r].key) swap(l, r);
      int lc = node[r].next = node[l].child;
      node[l].child = node[lc].prev = r;
      return node[r].prev = l;
    }

    int two_pass_pairing(int root) {
      if(!root) return 0;
      int a = root;
      root = 0;
      while(a) {
        int b = node[a].next, na = 0;
        node[a].prev = node[a].next = 0;
        if(b) na = node[b].next, node[b].prev = node[b].next = 0;
        a = merge(a, b);
        node[a].next = root;
        root = a;
        a = na;
      }
      int s = node[root].next;
      node[root].next = 0;
      while(s) {
        int t = node[s].next;
        node[s].next = 0;
        root = merge(root, s);
        s = t;
      }
      return root;
    }

  private:
    vector< int > heap;
    vector< Node > node;
  };

  template< typename T >
  struct PriorityQueue : public priority_queue< T, vector< T >, greater< T > > {
    PriorityQueue() {}

    PriorityQueue(int N) { this->c.reserve(N); }

    T min() const { return this->top(); }

    void clear() { this->c.clear(); }
  };

  template< typename T >
  struct Queue {
    Queue() {}

    Queue(int N) : qh(0), qt(0), data(N) {}

    T operator[](int i) const { return data[i]; }

    void enqueue(int u) { data[qt++] = u; }

    int dequeue() { return data[qh++]; }

    bool empty() const { return qh == qt; }

    void clear() { qh = qt = 0; }

    int size() const { return qt; }

    int qh, qt;
    vector< T > data;
  };

public:
  struct InputEdge {
    int from, to;
    cost_t cost;
  };

private:
  template< typename T > using ModifiableHeap = BinaryHeap< T >;
  template< typename T > using ModifiableHeaps = PairingHeaps< T >;
  template< typename T > using FastHeap = PriorityQueue< T >;

  struct Edge {
    int to;
    cost_t cost;
  };
  struct Link {
    int from, to;
  };

  struct Node {
    struct NodeLink {
      int b, v;
    };

    Node() {}

    Node(int u) : parent(0), size(1) { link[0] = link[1] = {u, u}; }

    int next_v() const { return link[0].v; }

    int next_b() const { return link[0].b; }

    int prev_v() const { return link[1].v; }

    int prev_b() const { return link[1].b; }

    int parent, size;
    NodeLink link[2];
  };

  struct Event {
    Event() {}

    Event(cost_t time, int id) : time(time), id(id) {}

    bool operator<(const Event &rhs) const { return time < rhs.time; }

    bool operator>(const Event &rhs) const { return time > rhs.time; }

    cost_t time;
    int id;
  };

  struct EdgeEvent {
    EdgeEvent() {}

    EdgeEvent(cost_t time, int from, int to) : time(time), from(from), to(to) {}

    bool operator>(const EdgeEvent &rhs) const { return time > rhs.time; }

    bool operator<(const EdgeEvent &rhs) const { return time < rhs.time; }

    cost_t time;
    int from, to;
  };

public:
  MaximumWeightedMatching(int N, const vector< InputEdge > &in)
      : N(N), B((N - 1) / 2), S(N + B + 1), ofs(N + 2), edges(in.size() * 2),
        heap2(S), heap2s(S, S), heap3(edges.size()), heap4(S) {

    for(auto &e : in) ofs[e.from + 1]++, ofs[e.to + 1]++;
    for(int i = 1; i <= N + 1; ++i) ofs[i] += ofs[i - 1];
    for(auto &e : in) {
      edges[ofs[e.from]++] = {e.to, e.cost * 2};
      edges[ofs[e.to]++] = {e.from, e.cost * 2};
    }
    for(int i = N + 1; i > 0; --i) ofs[i] = ofs[i - 1];
    ofs[0] = 0;
  }

  tcost_t maximum_weighted_matching(bool init_matching = false) {
    initialize();
    set_potential();
    if(init_matching) find_maximal_matching();
    for(int u = 1; u <= N; ++u) if(!mate[u]) do_edmonds_search(u);
    tcost_t ret = compute_optimal_value();
    return ret;
  }

private:
  tcost_t compute_optimal_value() const {
    tcost_t ret = 0;
    for(int u = 1; u <= N; ++u)
      if(mate[u] > u) {
        cost_t max_c = 0;
        for(int eid = ofs[u]; eid < ofs[u + 1]; ++eid) {
          if(edges[eid].to == mate[u]) max_c = max(max_c, edges[eid].cost);
        }
        ret += max_c;
      }
    return ret >> 1;
  }

  inline tcost_t reduced_cost(int u, int v, const Edge &e) const {
    return tcost_t(potential[u]) + potential[v] - e.cost;
  }

  void rematch(int v, int w) {
    int t = mate[v];
    mate[v] = w;
    if(mate[t] != v) return;
    if(link[v].to == surface[link[v].to]) {
      mate[t] = link[v].from;
      rematch(mate[t], t);
    } else {
      int x = link[v].from, y = link[v].to;
      rematch(x, y);
      rematch(y, x);
    }
  }

  void fix_mate_and_base(int b) {
    if(b <= N) return;
    int bv = base[b], mv = node[bv].link[0].v, bmv = node[bv].link[0].b;
    int d = (node[bmv].link[1].v == mate[mv]) ? 0 : 1;
    while(1) {
      int mv = node[bv].link[d].v, bmv = node[bv].link[d].b;
      if(node[bmv].link[1 ^ d].v != mate[mv]) break;
      fix_mate_and_base(bv);
      fix_mate_and_base(bmv);
      bv = node[bmv].link[d].b;
    }
    fix_mate_and_base(base[b] = bv);
    mate[b] = mate[bv];
  }

  void reset_time() {
    time_current_ = 0;
    event1 = {Inf, 0};
  }

  void reset_blossom(int b) {
    label[b] = kFree;
    link[b].from = 0;
    slack[b] = Inf;
    lazy[b] = 0;
  }

  void reset_all() {
    label[0] = kFree;
    link[0].from = 0;
    for(int v = 1; v <= N; ++v) { // should be optimized for sparse graphs.
      if(label[v] == kOuter) potential[v] -= time_current_;
      else {
        int bv = surface[v];
        potential[v] += lazy[bv];
        if(label[bv] == kInner) potential[v] += time_current_ - time_created[bv];
      }
      reset_blossom(v);
    }
    for(int b = N + 1, r = B - unused_bid_idx_; r > 0 && b < S; ++b)
      if(base[b] != b) {
        if(surface[b] == b) {
          fix_mate_and_base(b);
          if(label[b] == kOuter) potential[b] += (time_current_ - time_created[b]) << 1;
          else if(label[b] == kInner) fix_blossom_potential< kInner >(b);
          else fix_blossom_potential< kFree >(b);
        }
        heap2s.clear(b);
        reset_blossom(b);
        --r;
      }

    que.clear();
    reset_time();
    heap2.clear();
    heap3.clear();
    heap4.clear();
  }

  void do_edmonds_search(int root) {
    if(potential[root] == 0) return;
    link_blossom(surface[root], {0, 0});
    push_outer_and_fix_potentials(surface[root], 0);
    for(bool augmented = false; !augmented;) {
      augmented = augment(root);
      if(augmented) break;
      augmented = adjust_dual_variables(root);
    }
    reset_all();
  }

  template< Label Lab >
  inline cost_t fix_blossom_potential(int b) {
    // Return the amount.
    // (If v is an atom, the potential[v] will not be changed.)
    cost_t d = lazy[b];
    lazy[b] = 0;
    if(Lab == kInner) {
      cost_t dt = time_current_ - time_created[b];
      if(b > N) potential[b] -= dt << 1;
      d += dt;
    }
    return d;
  }

  template< Label Lab >
  inline void update_heap2(int x, int y, int by, cost_t t) {
    if(t >= slack[y]) return;
    slack[y] = t;
    best_from[y] = x;
    if(y == by) {
      if(Lab != kInner) heap2.decrease_key(y, EdgeEvent(t + lazy[y], x, y));
    } else {
      int gy = group[y];
      if(gy != y) {
        if(t >= slack[gy]) return;
        slack[gy] = t;
      }
      heap2s.decrease_key(by, gy, EdgeEvent(t, x, y));
      if(Lab == kInner) return;
      EdgeEvent m = heap2s.min(by);
      heap2.decrease_key(by, EdgeEvent(m.time + lazy[by], m.from, m.to));
    }
  }

  void activate_heap2_node(int b) {
    if(b <= N) {
      if(slack[b] < Inf) heap2.push(b, EdgeEvent(slack[b] + lazy[b], best_from[b], b));
    } else {
      if(heap2s.empty(b)) return;
      EdgeEvent m = heap2s.min(b);
      heap2.push(b, EdgeEvent(m.time + lazy[b], m.from, m.to));
    }
  }

  void swap_blossom(int a, int b) {
    // Assume that `b` is a maximal blossom.
    swap(base[a], base[b]);
    if(base[a] == a) base[a] = b;
    swap(heavy[a], heavy[b]);
    if(heavy[a] == a) heavy[a] = b;
    swap(link[a], link[b]);
    swap(mate[a], mate[b]);
    swap(potential[a], potential[b]);
    swap(lazy[a], lazy[b]);
    swap(time_created[a], time_created[b]);
    for(int d = 0; d < 2; ++d) node[node[a].link[d].b].link[1 ^ d].b = b;
    swap(node[a], node[b]);
  }

  void set_surface_and_group(int b, int sf, int g) {
    surface[b] = sf, group[b] = g;
    if(b <= N) return;
    for(int bb = base[b]; surface[bb] != sf; bb = node[bb].next_b()) {
      set_surface_and_group(bb, sf, g);
    }
  }

  void merge_smaller_blossoms(int bid) {
    int lb = bid, largest_size = 1;
    for(int beta = base[bid], b = beta;;) {
      if(node[b].size > largest_size) largest_size = node[b].size, lb = b;
      if((b = node[b].next_b()) == beta) break;
    }
    for(int beta = base[bid], b = beta;;) {
      if(b != lb) set_surface_and_group(b, lb, b);
      if((b = node[b].next_b()) == beta) break;
    }
    group[lb] = lb;
    if(largest_size > 1) {
      surface[bid] = heavy[bid] = lb;
      swap_blossom(lb, bid);
    } else heavy[bid] = 0;
  }

  void contract(int x, int y, int eid) {
    int bx = surface[x], by = surface[y];
    assert(bx != by);
    const int h = -(eid + 1);
    link[surface[mate[bx]]].from = link[surface[mate[by]]].from = h;

    int lca = -1;
    while(1) {
      if(mate[by] != 0) swap(bx, by);
      bx = lca = surface[link[bx].from];
      if(link[surface[mate[bx]]].from == h) break;
      link[surface[mate[bx]]].from = h;
    }

    const int bid = unused_bid[--unused_bid_idx_];
    assert(unused_bid_idx_ >= 0);
    int tree_size = 0;
    for(int d = 0; d < 2; ++d) {
      for(int bv = surface[x]; bv != lca;) {
        int mv = mate[bv], bmv = surface[mv], v = mate[mv];
        int f = link[v].from, t = link[v].to;
        tree_size += node[bv].size + node[bmv].size;
        link[mv] = {x, y};

        if(bv > N) potential[bv] += (time_current_ - time_created[bv]) << 1;
        if(bmv > N) heap4.erase(bmv);
        push_outer_and_fix_potentials(bmv, fix_blossom_potential< kInner >(bmv));

        node[bv].link[d] = {bmv, mv};
        node[bmv].link[1 ^ d] = {bv, v};
        node[bmv].link[d] = {bv = surface[f], f};
        node[bv].link[1 ^ d] = {bmv, t};
      }
      node[surface[x]].link[1 ^ d] = {surface[y], y};
      swap(x, y);
    }
    if(lca > N) potential[lca] += (time_current_ - time_created[lca]) << 1;
    node[bid].size = tree_size + node[lca].size;
    base[bid] = lca;
    link[bid] = link[lca];
    mate[bid] = mate[lca];
    label[bid] = kOuter;
    surface[bid] = bid;
    time_created[bid] = time_current_;
    potential[bid] = 0;
    lazy[bid] = 0;

    merge_smaller_blossoms(bid); // O(n log n) time / Edmonds search
  }

  void link_blossom(int v, Link l) {
    link[v] = {l.from, l.to};
    if(v <= N) return;
    int b = base[v];
    link_blossom(b, l);
    int pb = node[b].prev_b();
    l = {node[pb].next_v(), node[b].prev_v()};
    for(int bv = b;;) {
      int bw = node[bv].next_b();
      if(bw == b) break;
      link_blossom(bw, l);
      Link nl = {node[bw].prev_v(), node[bv].next_v()};
      bv = node[bw].next_b();
      link_blossom(bv, nl);
    }
  }

  void push_outer_and_fix_potentials(int v, cost_t d) {
    label[v] = kOuter;
    if(v > N) {
      for(int b = base[v]; label[b] != kOuter; b = node[b].next_b()) {
        push_outer_and_fix_potentials(b, d);
      }
    } else {
      potential[v] += time_current_ + d;
      if(potential[v] < event1.time) event1 = {potential[v], v};
      que.enqueue(v);
    }
  }

  bool grow(int x, int y) {
    int by = surface[y];
    bool visited = (label[by] != kFree);
    if(!visited) link_blossom(by, {0, 0});
    label[by] = kInner;
    time_created[by] = time_current_;
    heap2.erase(by);
    if(y != by) heap4.update(by, time_current_ + (potential[by] >> 1));
    int z = mate[by];
    if(z == 0) {
      rematch(x, y);
      rematch(y, x);
      return true;
    }
    int bz = surface[z];
    if(!visited) link_blossom(bz, {x, y});
    else link[bz] = link[z] = {x, y};
    push_outer_and_fix_potentials(bz, fix_blossom_potential< kFree >(bz));
    time_created[bz] = time_current_;
    heap2.erase(bz);
    return false;
  }

  void free_blossom(int bid) {
    unused_bid[unused_bid_idx_++] = bid;
    base[bid] = bid;
  }

  int recalculate_minimum_slack(int b, int g) {
    // Return the destination of the best edge of blossom `g`.
    if(b <= N) {
      if(slack[b] >= slack[g]) return 0;
      slack[g] = slack[b];
      best_from[g] = best_from[b];
      return b;
    }
    int v = 0;
    for(int beta = base[b], bb = beta;;) {
      int w = recalculate_minimum_slack(bb, g);
      if(w != 0) v = w;
      if((bb = node[bb].next_b()) == beta) break;
    }
    return v;
  }

  void construct_smaller_components(int b, int sf, int g) {
    surface[b] = sf, group[b] = g; // `group[b] = g` is unneeded.
    if(b <= N) return;
    for(int bb = base[b]; surface[bb] != sf; bb = node[bb].next_b()) {
      if(bb == heavy[b]) {
        construct_smaller_components(bb, sf, g);
      } else {
        set_surface_and_group(bb, sf, bb);
        int to = 0;
        if(bb > N) slack[bb] = Inf, to = recalculate_minimum_slack(bb, bb);
        else if(slack[bb] < Inf) to = bb;
        if(to > 0) heap2s.push(sf, bb, EdgeEvent(slack[bb], best_from[bb], to));
      }
    }
  }

  void move_to_largest_blossom(int bid) {
    const int h = heavy[bid];
    cost_t d = (time_current_ - time_created[bid]) + lazy[bid];
    lazy[bid] = 0;
    for(int beta = base[bid], b = beta;;) {
      time_created[b] = time_current_;
      lazy[b] = d;
      if(b != h) construct_smaller_components(b, b, b), heap2s.erase(bid, b);
      if((b = node[b].next_b()) == beta) break;
    }
    if(h > 0) swap_blossom(h, bid), bid = h;
    free_blossom(bid);
  }

  void expand(int bid) {
    int mv = mate[base[bid]];
    move_to_largest_blossom(bid); // O(n log n) time / Edmonds search
    Link old_link = link[mv];
    int old_base = surface[mate[mv]], root = surface[old_link.to];
    int d = (mate[root] == node[root].link[0].v) ? 1 : 0;
    for(int b = node[old_base].link[d ^ 1].b; b != root;) {
      label[b] = kSeparated;
      activate_heap2_node(b);
      b = node[b].link[d ^ 1].b;
      label[b] = kSeparated;
      activate_heap2_node(b);
      b = node[b].link[d ^ 1].b;
    }
    for(int b = old_base;; b = node[b].link[d].b) {
      label[b] = kInner;
      int nb = node[b].link[d].b;
      if(b == root) link[mate[b]] = old_link;
      else link[mate[b]] = {node[b].link[d].v, node[nb].link[d ^ 1].v};
      link[surface[mate[b]]] = link[mate[b]]; // fix tree links
      if(b > N) {
        if(potential[b] == 0) expand(b);
        else heap4.push(b, time_current_ + (potential[b] >> 1));
      }
      if(b == root) break;
      push_outer_and_fix_potentials(nb, fix_blossom_potential< kInner >(b = nb));
    }
  }

  bool augment(int root) {
    // Return true if an augmenting path is found.
    while(!que.empty()) {
      int x = que.dequeue(), bx = surface[x];
      if(potential[x] == time_current_) {
        if(x != root) rematch(x, 0);
        return true;
      }
      for(int eid = ofs[x]; eid < ofs[x + 1]; ++eid) {
        auto &e = edges[eid];
        int y = e.to, by = surface[y];
        if(bx == by) continue;
        Label l = label[by];
        if(l == kOuter) {
          cost_t t = reduced_cost(x, y, e) >> 1; // < 2 * Inf
          if(t == time_current_) {
            contract(x, y, eid);
            bx = surface[x];
          } else if(t < event1.time) {
            heap3.emplace(t, x, eid);
          }
        } else {
          tcost_t t = reduced_cost(x, y, e); // < 3 * Inf
          if(t >= Inf) continue;
          if(l != kInner) {
            if(cost_t(t) + lazy[by] == time_current_) {
              if(grow(x, y)) return true;
            } else update_heap2< kFree >(x, y, by, t);
          } else {
            if(mate[x] != y) update_heap2< kInner >(x, y, by, t);
          }
        }
      }
    }
    return false;
  }

  bool adjust_dual_variables(int root) {
    // delta1 : rematch
    cost_t time1 = event1.time;

    // delta2 : grow
    cost_t time2 = Inf;
    if(!heap2.empty()) time2 = heap2.min().time;

    // delta3 : contract : O(m log n) time / Edmonds search [ bottleneck (?) ]
    cost_t time3 = Inf;
    while(!heap3.empty()) {
      EdgeEvent e = heap3.min();
      int x = e.from, y = edges[e.to].to; // e.to is some edge id.
      if(surface[x] != surface[y]) {
        time3 = e.time;
        break;
      } else heap3.pop();
    }

    // delta4 : expand
    cost_t time4 = Inf;
    if(!heap4.empty()) time4 = heap4.min();

    // -- events --
    cost_t time_next = min(min(time1, time2), min(time3, time4));
    assert(time_current_ <= time_next && time_next < Inf);
    time_current_ = time_next;

    if(time_current_ == event1.time) {
      int x = event1.id;
      if(x != root) rematch(x, 0);
      return true;
    }
    while(!heap2.empty() && heap2.min().time == time_current_) {
      int x = heap2.min().from, y = heap2.min().to;
      if(grow(x, y)) return true; // `grow` function will call `heap2.erase(by)`.
    }
    while(!heap3.empty() && heap3.min().time == time_current_) {
      int x = heap3.min().from, eid = heap3.min().to;
      int y = edges[eid].to;
      heap3.pop();
      if(surface[x] == surface[y]) continue;
      contract(x, y, eid);
    }
    while(!heap4.empty() && heap4.min() == time_current_) {
      int b = heap4.argmin();
      heap4.pop();
      expand(b);
    }
    return false;
  }

private:
  void initialize() {
    que = Queue< int >(N);
    mate.assign(S, 0);
    link.assign(S, {0, 0});
    label.assign(S, kFree);
    base.resize(S);
    for(int u = 1; u < S; ++u) base[u] = u;
    surface.resize(S);
    for(int u = 1; u < S; ++u) surface[u] = u;

    potential.resize(S);
    node.resize(S);
    for(int b = 1; b < S; ++b) node[b] = Node(b);

    unused_bid.resize(B);
    for(int i = 0; i < B; ++i) unused_bid[i] = N + B - i;
    unused_bid_idx_ = B;

    // for O(nm log n) implementation
    reset_time();
    time_created.resize(S);
    slack.resize(S);
    for(int i = 0; i < S; ++i) slack[i] = Inf;
    best_from.assign(S, 0);
    heavy.assign(S, 0);
    lazy.assign(S, 0);
    group.resize(S);
    for(int i = 0; i < S; ++i) group[i] = i;
  }

  void set_potential() {
    for(int u = 1; u <= N; ++u) {
      cost_t max_c = 0;
      for(int eid = ofs[u]; eid < ofs[u + 1]; ++eid) {
        max_c = max(max_c, edges[eid].cost);
      }
      potential[u] = max_c >> 1;
    }
  }

  void find_maximal_matching() {
    // Find a maximal matching naively.
    for(int u = 1; u <= N; ++u)
      if(!mate[u]) {
        for(int eid = ofs[u]; eid < ofs[u + 1]; ++eid) {
          auto &e = edges[eid];
          int v = e.to;
          if(mate[v] > 0 || reduced_cost(u, v, e) > 0) continue;
          mate[u] = v;
          mate[v] = u;
          break;
        }
      }
  }

private:
  const int N, B, S; // N = |V|, B = (|V| - 1) / 2, S = N + B + 1
  vector< int > ofs;
  vector< Edge > edges;

  Queue< int > que;
  vector< int > mate, surface, base;
  vector< Link > link;
  vector< Label > label;
  vector< cost_t > potential;

  vector< int > unused_bid;
  int unused_bid_idx_;
  vector< Node > node;

  // for O(nm log n) implementation
  vector< int > heavy, group;
  vector< cost_t > time_created, lazy, slack;
  vector< int > best_from;

  cost_t time_current_;
  Event event1;
  ModifiableHeap< EdgeEvent > heap2;
  ModifiableHeaps< EdgeEvent > heap2s;
  FastHeap< EdgeEvent > heap3;
  ModifiableHeap< cost_t > heap4;
};

using MWM = MaximumWeightedMatching< int >;
using Edge = MWM::InputEdge;


int main() {
  int N, V[24][24];
  cin >> N;
  vector< Edge > edges;
  for(int i = 0; i < N; i++) {
    for(int j = 0; j < N; j++) {
      cin >> V[i][j];
    }
    for(int j = i + 1; j < N; j++) {
      edges.push_back({i + 1, j + 1, V[i][j]});
    }
  }
  MWM flow(N, edges);
  cout << flow.maximum_weighted_matching() << endl;
}