先决条件： Fenwick树(二进制索引树)
给定一个由N个数字组成的数组，以及多个查询，其中每个查询将包含三个数字(l，r和k)。任务是计算子数组[L，R]中大于K的数组元素的数量。
例子：
Input: n=6
q=2
arr[ ] = { 7, 3, 9, 13, 5, 4 }
Query1: l=1, r=4, k=6
Query2: l=2, r=6, k=8
Output: 3
2
For the first query, [7, 3, 9, 13] represents the
subarray from index 1 till 4, in which there are
3 numbers which are greater than k=6 that are {7, 9, 13}.
For the second query, there are only
two numbers in the query range which
are greater than k.
天真的方法是通过简单地从索引l遍历到r遍历数组来查找每个查询的答案，并在数组元素大于k时保持对计数加1。
时间复杂度： O(n * q)
更好的方法是使用合并排序树。用这种方法，在每个节点上建立一个带有向量的分段树，该向量包含按排序顺序包含子范围的所有元素。使用段树回答每个查询，其中可使用二元搜索来计算每个子范围内的子节点位于查询范围内的数目大于k的数目。
时间复杂度： O(q * log(n)* log(n))
一种有效的方法是使用脱机查询和Fenwick树来解决问题。步骤如下：
- 首先将所有数组元素和查询存储在同一数组中。为此，我们可以创建一个自结构或类。
- 然后按降序对结构数组进行排序(如果发生冲突，查询将首先出现，然后是数组元素)。
- 再次处理整个结构数组，但是在此之前，创建另一个BIT数组(二叉索引树)，其query(i)函数将返回数组中存在的所有元素的计数，直到第i个索引为止。
- 最初，用0填充整个数组。
- 创建一个答案数组，其中存储每个查询的答案。
- 处理结构数组。
- 如果它是一个数组元素，则从该元素的索引中用+1更新BIT数组。
- 如果是查询，则减去query(r)– query(l-1) ，这将是该查询的答案，该答案将存储在答案数组中与查询编号相对应的索引处。
- 最后输出答案数组。
此处的主要观察结果是，由于结构的数组已按降序排序。每当我们遇到任何查询时，只有大于“ k”的元素才包含在BIT数组中的计数，这是需要的答案。
下面是程序中使用的结构的说明：
Pos: stores the order of query. In case of array elements it is kept as 0.
L: stores the starting index of the query’s subarray. In case of array elements it is also 0.
R: stores the ending idex of the query’s subarray. In case of array element it is used to store the position of element in the array.
Val: store ‘k’ of the query and all the array elements.
下面是上述方法的实现：
C++
// C++ program to print the number of elements
// greater than k in a subarray of range L-R.
#include
using namespace std;
// Structure which will store both
// array elements and queries.
struct node {
int pos;
int l;
int r;
int val;
};
// Boolean comparator that will be used
// for sorting the structural array.
bool comp(node a, node b)
{
// If 2 values are equal the query will
// occur first then array element
if (a.val == b.val)
return a.l > b.l;
// Otherwise sorted in descending order.
return a.val > b.val;
}
// Updates the node of BIT array by adding
// 1 to it and its ancestors.
void update(int* BIT, int n, int idx)
{
while (idx <= n) {
BIT[idx]++;
idx += idx & (-idx);
}
}
// Returns the count of numbers of elements
// present from starting till idx.
int query(int* BIT, int idx)
{
int ans = 0;
while (idx) {
ans += BIT[idx];
idx -= idx & (-idx);
}
return ans;
}
// Function to solve the queries offline
void solveQuery(int arr[], int n, int QueryL[],
int QueryR[], int QueryK[], int q)
{
// create node to store the elements
// and the queries
node a[n + q + 1];
// 1-based indexing.
// traverse for all array numbers
for (int i = 1; i <= n; ++i) {
a[i].val = arr[i - 1];
a[i].pos = 0;
a[i].l = 0;
a[i].r = i;
}
// iterate for all queries
for (int i = n + 1; i <= n + q; ++i) {
a[i].pos = i - n;
a[i].val = QueryK[i - n - 1];
a[i].l = QueryL[i - n - 1];
a[i].r = QueryR[i - n - 1];
}
// In-built sort function used to
// sort node array using comp function.
sort(a + 1, a + n + q + 1, comp);
// Binary Indexed tree with
// initially 0 at all places.
int BIT[n + 1];
// initially 0
memset(BIT, 0, sizeof(BIT));
// For storing answers for each query( 1-based indexing ).
int ans[q + 1];
// traverse for numbers and query
for (int i = 1; i <= n + q; ++i) {
if (a[i].pos != 0) {
// call function to returns answer for each query
int cnt = query(BIT, a[i].r) - query(BIT, a[i].l - 1);
// This will ensure that answer of each query
// are stored in order it was initially asked.
ans[a[i].pos] = cnt;
}
else {
// a[i].r contains the position of the
// element in the original array.
update(BIT, n, a[i].r);
}
}
// Output the answer array
for (int i = 1; i <= q; ++i) {
cout << ans[i] << endl;
}
}
// Driver Code
int main()
{
int arr[] = { 7, 3, 9, 13, 5, 4 };
int n = sizeof(arr) / sizeof(arr[0]);
// 1-based indexing
int QueryL[] = { 1, 2 };
int QueryR[] = { 4, 6 };
// k for each query
int QueryK[] = { 6, 8 };
// number of queries
int q = sizeof(QueryL) / sizeof(QueryL[0]);
// Function call to get
solveQuery(arr, n, QueryL, QueryR, QueryK, q);
return 0;
} Python3
# Python program to print the number of elements
# greater than k in a subarray of range L-R.
# Structure which will store both
# array elements and queries.
class node:
def __init__(self):
self.pos = 0
self.l = 0
self.r = 0
self.val = 0
# Updates the node of BIT array by adding
# 1 to it and its ancestors.
def update(BIT: list, n: int, idx: int):
while idx <= n:
BIT[idx] += 1
idx += idx & -idx
# Returns the count of numbers of elements
# present from starting till idx.
def query(BIT: list, idx: int) -> int:
ans = 0
while idx:
ans += BIT[idx]
idx -= idx & -idx
return ans
# Function to solve the queries offline
def solveQuery(arr: list, n: int, QueryL: list,
QueryR: list, QueryK: list, q: int):
# create node to store the elements
# and the queries
a = [0] * (n + q + 1)
for i in range(n + q + 1):
a[i] = node()
# 1-based indexing
# traverse for all array numbers
for i in range(1, n + 1):
a[i].val = arr[i - 1]
a[i].pos = 0
a[i].l = 0
a[i].r = i
# iterate for all queries
for i in range(n + 1, n + q + 1):
a[i].pos = i - n
a[i].val = QueryK[i - n - 1]
a[i].l = QueryL[i - n - 1]
a[i].r = QueryR[i - n - 1]
# In-built sorted function used to
# sort node array using comp function.
a = [a[0]] + sorted(a[1:],
key = lambda k: (k.val, k.l),
reverse = True)
# Binary Indexed tree with
# initially 0 at all places.
BIT = [0] * (n + 1)
# For storing answers for
# each query( 1-based indexing ).
ans = [0] * (q + 1)
# traverse for numbers and query
for i in range(1, n + q + 1):
if a[i].pos != 0:
# call function to returns answer for each query
cnt = query(BIT, a[i].r) - query(BIT, a[i].l - 1)
# This will ensure that answer of each query
# are stored in order it was initially asked.
ans[a[i].pos] = cnt
else:
# a[i].r contains the position of the
# element in the original array.
update(BIT, n, a[i].r)
# Output the answer array
for i in range(1, q + 1):
print(ans[i])
# Driver Code
if __name__ == "__main__":
arr = [7, 3, 9, 13, 5, 4]
n = len(arr)
# 1-based indexing
QueryL = [1, 2]
QueryR = [4, 6]
# k for each query
QueryK = [6, 8]
# number of queries
q = len(QueryL)
# Function call to get
solveQuery(arr, n, QueryL, QueryR, QueryK, q)
# This code is contributed by
# sanjeev2552输出：
3
2
时间复杂度： O(N * log N)，其中N =(n + q)
什么是离线查询?
在某些问题中，很难以任何随机顺序回答查询。因此，与其单独回答每个查询，不如存储所有查询，然后对它们进行相应排序以有效地为它们计算答案。存储所有答案，然后按照最初给出的顺序输出。
此技术称为“脱机查询” 。
注意：代替树形树，也可以使用分段树，其中分段树的每个节点将存储在该迭代之前插入的元素数。更新和查询功能将更改，其余实现将保持不变。
脱机查询的必要条件：仅当一个查询的答案不依赖于先前查询的答案时，才可以使用此技术，因为排序后查询的顺序可能会更改。