看的网上的，不甚理解。代码中有注释。维护树状数组C[i]， sum(x)指 A[1:x]中有几个不同的数字
\
DQUERY - D-query
\
| English | Vietnamese |
|---|
Given a sequence of n numbers a1, a2, ..., an and a number of d-queries. A d-query is a pair (i, j) (1 ≤ i ≤ j ≤ n). For each d-query (i, j), you have to return the number of distinct elements in the subsequence ai, ai+1, ..., aj.
Input
- Line 1: n (1 ≤ n ≤ 30000).
- Line 2: n numbers a1, a2, ..., an (1 ≤ ai ≤ 106).
- Line 3: q (1 ≤ q ≤ 200000), the number of d-queries.
- In the next q lines, each line contains 2 numbers i, j representing a d-query (1 ≤ i ≤ j ≤ n).
Output
- For each d-query (i, j), print the number of distinct elements in the subsequence ai, ai+1, ..., aj in a single line.
Example
Input
5
1 1 2 1 3
3
1 5
2 4
3 5
Output
3
2
3
//树状数组离线操作
#include <cstdio>
#include <cstring>
#include <algorithm>
#include <map>
#define M 200007
using namespace std;
//cnt[i]存第i次询问结果，l,r分别存询问的边界，A,C是树状数组里边的
const int N = 3e4 + 7;
int l[M], r[M], A[N], id[M], n; //pos[i]指数字i在A[]中的位置，拖i在A中重复出现，维护pos[i]尽量靠后
int C[N], cnt[M], pos[1000007]; //pos的大小取决于A[i]的范围，可以换为数组
inline int lowbit(int x) {
return x & (-x);
}
void add(int x, int val) {
for (int i = x; i <= n; i += lowbit(i) )
C[i] += val;
}
int sum(int x) {
int res = 0;
for (int i = x; i > 0; i -= lowbit(i))
res += C[i];
return res;
}
bool cmp(int i, int j) {
return r[i] < r[j];
}
int main()
{
while(~scanf("%d", &n)) {
memset(C, 0, sizeof C); //树状数组
memset(pos, 0, sizeof pos);
for (int i = 1; i <= n; ++i) {
scanf("%d", A + i); //一维数组存数，和算法入门经典里讲的一样
add(i, 1); //
if (!pos[A[i]])
pos[A[i]] = i;
int m;
scanf("%d", &m);
for (int i = 1; i <= m; ++i) {
scanf("%d%d", l + i, r + i);
id[i] = i; //记下查询顺序
}
//接下来根据查询的右边界对查询顺序排序
sort(id + 1, id + 1 + m, cmp); //第一次见这种操作是看acdart直播时
/*主要思想是将计数尽可能拖到后面来*/
int right = 1;
for (int i = 1; i <= m; ++i) {
int x = id[i];
while(right <= r[x]) {
if (pos[A[right] != right]) {
add(pos[A[right]], -1); //在后面出现了，前面的计数抵消。在后面计数，保证查询区间里的数不遗漏
pos[A[right]] = right;
}
++right;
}
right = r[x];
cnt[x] = sum(r[x]) - sum(l[x] - 1);
}
for (int i = 1; i <= m; ++i)
printf("%d\n", cnt[i]);
}
}
return 0;
}
\
杭电hdu4417也可以用树状数组离线操作解决
#include <cstdio>
#include <cstring>
#include <algorithm>
#define N 100007
int bit[N], ary[N], cnt[N], n;
struct Node {
int val, pos;
bool operator < (const Node tmp) const {
return val < tmp.val;
}
}a[N];
struct Query {
int le, ri, id, h;
bool operator < (const Query tmp) const {
return h < tmp.h;
}
}q[N];
inline int lowbit(int x) {
return x & -x;
}
void add(int pos, int val) {
for(int i = pos; i <= n; i += lowbit(i))
bit[i] += val;
}
int sum(int pos) {
int ret = 0;
for (int i = pos; i > 0; i -= lowbit(i))
ret += bit[i];
return ret;
}
int main()
{
int T, kase = 0;
scanf("%d", &T);
while(T-- > 0) {
memset(bit, 0, sizeof bit);
int m;
scanf("%d%d", &n, &m);
for (int i= 1; i <= n; ++i) {
scanf("%d", &a[i].val);
a[i].pos = i;
}
for (int i = 1; i <= m; ++i) {
scanf("%d%d%d", &q[i].le, &q[i].ri, &q[i].h);
q[i].id = i;
}
std::sort(a + 1, a + 1 + n);
std::sort(q + 1, q + 1 + m);
int k = 1;
for (int i = 1; i <= m; ++i) {
while(k <= n && a[k].val <= q[i].h) {
add(a[k].pos, 1);
++k;
}
cnt[q[i].id] = sum(q[i].ri + 1) - sum(q[i].le);
}
printf("Case %d:\n", ++kase);
for (int i = 1; i <= m; ++i)
printf("%d\n", cnt[i]);
}
}
\
主席树解法
//求区间不同数的个数
#define _CRT_SECURE_NO_WARNINGS
#include <cstdio>
#include <algorithm>
#include <cstring>
#include <map>
using namespace std;
const int N = 30007; //最多N个数
const int M = N * 100;
int n, q, tot; //n个数，q次询问
int a[N]; //原始数组
//T[i]保存[1,i]中有多少个不同的数
int T[M], lson[M], rson[M], c[M]; //c数组可能就是主席树要维护的值，比如本题维护区间数的个数
int build(int l, int r) { //返回[l,r]的父节点。建一棵非叶子节点都有两个孩子的树
int rt = tot++; //tot从0开始，用的节点数。最终tot=2*n-1
c[rt] = 0;
if (l != r) {
int mid = l + r >> 1;
lson[rt] = build(l, mid); //lson用来保存父亲节点rt的左孩子的索引
rson[rt] = build(mid + 1, r);
}
return rt;
}
int update(int rt, int pos, int val) {
int newRoot = tot++, tmp = newRoot; //newRoot申请新的节点
c[newRoot] = c[rt] + val;//主席树可以加减。同构
int l = 1, r = n;
while (l < r) {//二分建树，只是写法形式上和 mid = l + r >> 1然后左右建树不同
int mid = l + r >> 1;
if (pos <= mid) {
lson[newRoot] = tot++;//新建左子树
rson[newRoot] = rson[rt];//右子树复用前一棵线段树的，这样就节省了空间，不至于MLE
newRoot = lson[newRoot];
rt = lson[rt];
r = mid;
}
else {
rson[newRoot] = tot++;
lson[newRoot] = lson[rt];
newRoot = rson[newRoot]; //上句要用到，newRoot，so,不能慌着更新newRoot
rt = rson[rt];//更新当前根节点
l = mid + 1;
}
c[newRoot] = c[rt] + val;//新节点的值，由之前的节点更新而来
}
return tmp; //返回新建树的根节点
}
//询问l到pos有多少个不同的数, rt = T[l]
int query(int rt, int pos) {
int res = 0;
int l = 1, r = n;
while (pos < r) {
int mid = l + r >> 1;
if (pos <= mid) {//在左子树中查
r = mid;
rt = lson[rt];
}
else { //在右子树中搜
res += c[lson[rt]];
rt = rson[rt];
l = mid + 1;
}
}
return res + c[rt];
}
int main()
{
while (~scanf("%d", &n)) {
tot = 0;
for (int i = 1; i <= n; ++i)
scanf("%d", a + i);
T[n + 1] = build(1, n); //难道是先建一棵树
map<int, int> mp; //主席树是函数式编程，可能需要映射吧
for (int i = n; i >= 1; --i) {
if (mp.find(a[i]) == mp.end()) {
T[i] = update(T[i + 1], i, 1); //每次调用都新建一棵线段树。update返回的是新建树的根节点
}
else {
int tmp = update(T[i + 1], mp[a[i]], -1); //若a[i]不是很大，可用数组代替映射
T[i] = update(tmp, i, 1);
}
mp[a[i]] = i; //若之前存在a[i]，那么a[i]将被覆盖了，第二值保存的是较小的i
}
scanf("%d", &q);
while (q--) {
int l, r;
scanf("%d%d", &l, &r);
printf("%d\n", query(T[l], r));//由l确定在那棵线段树中查询
}
}
return 0;
}
\
\
\
\
