#include <bits/stdc++.h>
using namespace std;
// Structure to hold (count, candidate speciality) for a given distance.
struct PairEntry {
int cnt;
int cand; // if cnt==1, then cand is the speciality of the unique city; otherwise, cand is -1.
};
// Combine two distributions A and B (element–wise) so that for each index i the new
// count is A[i].cnt + B[i].cnt and if that sum is 1 the candidate is the one that appears.
vector<PairEntry> combineDist(const vector<PairEntry>& A, const vector<PairEntry>& B) {
int n = max(A.size(), B.size());
vector<PairEntry> res(n, {0, -1});
for (int i = 0; i < n; i++){
int a_cnt = (i < (int)A.size() ? A[i].cnt : 0);
int b_cnt = (i < (int)B.size() ? B[i].cnt : 0);
int tot = a_cnt + b_cnt;
int cand = -1;
if(tot == 1) {
if(i < (int)A.size() && A[i].cnt == 1)
cand = A[i].cand;
else if(i < (int)B.size() && B[i].cnt == 1)
cand = B[i].cand;
}
res[i] = {tot, cand};
}
return res;
}
// Merge a distribution 'child' into 'base', but with a shift of 1 (because if we attach child’s subtree
// then every node in that subtree is one road farther from the parent).
void mergeShift(vector<PairEntry>& base, const vector<PairEntry>& child) {
int shift = 1;
int newSize = max((int)base.size(), (int)child.size() + shift);
base.resize(newSize, {0, -1});
for (int i = 0; i < (int)child.size(); i++){
int pos = i + shift;
int tot = base[pos].cnt + child[i].cnt;
int cand = -1;
if(tot == 1) {
if(base[pos].cnt == 1)
cand = base[pos].cand;
else
cand = child[i].cand;
}
base[pos].cnt = tot;
base[pos].cand = cand;
}
}
// Global variables.
int N, M;
vector<int> C; // Speciality for each city (0-indexed).
vector<vector<int>> adj; // Tree (adjacency list).
vector<vector<PairEntry>> dp; // dp[x] will store the distribution for the subtree rooted at x.
vector<int> ans; // ans[x] will be the final answer for city x.
// dfs1 computes dp[x] = distribution for the subtree rooted at x.
// At distance 0 we record city x itself.
void dfs1(int x, int parent) {
dp[x].clear();
dp[x].push_back({1, C[x]}); // distance 0: only x itself.
for (int v : adj[x]) {
if(v == parent) continue;
dfs1(v, x);
mergeShift(dp[x], dp[v]);
}
}
// dfs2 does the re–rooting. For node x we are given 'up' – the distribution of all nodes that are not in x’s subtree,
// as seen from x. Then the full distribution for x is F[x] = combineDist(up, dp[x]).
// From F[x] we can count the number of distances d>=1 that contain a unique city and then compute the number of distinct specialities.
void dfs2(int x, int parent, const vector<PairEntry>& up) {
// Full distribution F[x] = (up from parent's side) + (dp[x] from x’s subtree)
vector<PairEntry> full = combineDist(up, dp[x]);
// For every distance d>=1, if full[d].cnt==1 then the unique city at that distance contributes its speciality.
unordered_set<int> uniqSpecs;
for (int d = 1; d < (int)full.size(); d++){
if(full[d].cnt == 1)
uniqSpecs.insert(full[d].cand);
}
ans[x] = uniqSpecs.size();
// For re–rooting we want to compute for each child v of x a new up–distribution.
// Let children[] be the list of x’s children (neighbors except the parent).
vector<int> children;
for (int v : adj[x])
if(v != parent)
children.push_back(v);
int k = children.size();
// Build prefix and suffix arrays over the children.
// For each child, the “branch” distribution is dp[v] shifted by 1.
vector< vector<PairEntry> > pref(k+1), suf(k+1);
pref[0].clear(); // identity: an empty distribution (which we interpret as all zeros).
for (int i = 0; i < k; i++){
int v = children[i];
vector<PairEntry> branch;
if(dp[v].empty()){
branch.clear();
} else {
branch.resize(dp[v].size() + 1, {0, -1});
// Shift dp[v] by 1.
for (int j = 0; j < (int)dp[v].size(); j++){
branch[j+1] = dp[v][j];
}
}
pref[i+1] = combineDist(pref[i], branch);
}
suf[k].clear();
for (int i = k-1; i >= 0; i--){
int v = children[i];
vector<PairEntry> branch;
if(dp[v].empty()){
branch.clear();
} else {
branch.resize(dp[v].size() + 1, {0, -1});
for (int j = 0; j < (int)dp[v].size(); j++){
branch[j+1] = dp[v][j];
}
}
suf[i] = combineDist(branch, suf[i+1]);
}
// Now, for each child v we want to compute its up–distribution.
// The idea is that the parent's full distribution F[x] comes from:
// (i) x itself (at distance 0), plus
// (ii) contributions from all children.
// For child v, we remove its branch from F[x] and then “shift” the result by 1.
for (int i = 0; i < k; i++){
int v = children[i];
// x’s own contribution is a base distribution: one node (x) at distance 0.
vector<PairEntry> baseVec;
baseVec.push_back({1, C[x]});
// Combine contributions from siblings: merge (prefix from children[0..i-1] and suffix from children[i+1..k-1]).
vector<PairEntry> siblings = combineDist(pref[i], suf[i+1]);
// Combine x’s own contribution with siblings.
vector<PairEntry> R = combineDist(baseVec, siblings);
// Also, include the parent's up–distribution that was passed to x.
vector<PairEntry> newUp = combineDist(up, R);
// Now, shift newUp by 1 to get the up–distribution as seen from child v.
vector<PairEntry> shiftedNewUp;
if(newUp.empty()){
shiftedNewUp.clear();
} else {
shiftedNewUp.resize(newUp.size() + 1, {0, -1});
for (int j = 0; j < (int)newUp.size(); j++){
shiftedNewUp[j+1] = newUp[j];
}
}
dfs2(v, x, shiftedNewUp);
}
}
// Main reads the tree and specialities, computes dp and answers, then prints them.
int main(){
ios::sync_with_stdio(false);
cin.tie(nullptr);
cin >> N >> M;
adj.assign(N, vector<int>());
dp.assign(N, vector<PairEntry>());
ans.assign(N, 0);
C.resize(N);
// Read the N-1 roads.
for (int i = 0; i < N-1; i++){
int u, v;
cin >> u >> v;
u--; v--; // converting to 0-indexed.
adj[u].push_back(v);
adj[v].push_back(u);
}
// Read the speciality for each city.
for (int i = 0; i < N; i++){
cin >> C[i];
}
// First DFS: compute dp for each subtree.
dfs1(0, -1);
// For the root, there is no parent's contribution (up is empty).
vector<PairEntry> up;
up.clear();
dfs2(0, -1, up);
// Output the answer for each city.
for (int i = 0; i < N; i++){
cout << ans[i] << "\n";
}
return 0;
}
# | Verdict | Execution time | Memory | Grader output |
---|
Fetching results... |
# | Verdict | Execution time | Memory | Grader output |
---|
Fetching results... |
# | Verdict | Execution time | Memory | Grader output |
---|
Fetching results... |
# | Verdict | Execution time | Memory | Grader output |
---|
Fetching results... |