/**
* @file
* @brief Implementation of [Sparse
* Table](https://brilliant.org/wiki/sparse-table/) for `min()` function.
* @author [Mann Patel](https://github.com/manncodes)
* @details
* Sparse Table is a data structure, that allows answering range queries.
* It can answer most range queries in O(logn), but its true power is answering
* range minimum queries (or equivalent range maximum queries). For those
* queries it can compute the answer in O(1) time. The only drawback of this
* data structure is, that it can only be used on immutable arrays. This means,
* that the array cannot be changed between two queries.
*
* If any element in the array changes, the complete data structure has to be
* recomputed.
*
* @todo make stress tests.
*
* @warning
* This sparse table is made for `min(a1,a2,...an)` duplicate invariant
* function. This implementation can be changed to other functions like
* `gcd()`, `lcm()`, and `max()` by changing a few lines of code.
*/
#include <array> /// for std::array
#include <cassert> /// for assert
#include <iostream> /// for IO operations
/**
* @namespace data_structures
* @brief Data Structures algorithms
*/
namespace data_structures {
/**
* @namespace sparse_table
* @brief Functions for Implementation of [Sparse
* Table](https://brilliant.org/wiki/sparse-table/)
*/
namespace sparse_table {
/**
* @brief A struct to represent sparse table for `min()` as their invariant
* function, for the given array `A`. The answer to queries are stored in the
* array ST.
*/
constexpr uint32_t N = 12345; ///< the maximum size of the array.
constexpr uint8_t M = 14; ///< ceil(log2(N)).
struct Sparse_table {
size_t n = 0; ///< size of input array.
/** @warning check if `N` is not less than `n`. if so, manually increase the
* value of N */
std::array<int64_t, N> A = {}; ///< input array to perform RMQ.
std::array<std::array<int64_t, N>, M>
ST{}; ///< the sparse table storing `min()` values for given interval.
std::array<int64_t, N> LOG = {}; ///< where floor(log2(i)) are precomputed.
/**
* @brief Builds the sparse table for computing min/max/gcd/lcm/...etc
* for any contiguous sub-segment of the array.This is an example of
* computing the index of the minimum value.
* @return void
* @complexity: O(n.log(n))
*/
void buildST() {
LOG[0] = -1;
for (size_t i = 0; i < n; ++i) {
ST[0][i] = static_cast<int64_t>(i);
LOG[i + 1] = LOG[i] + !(i & (i + 1)); ///< precomputing `log2(i+1)`
}
for (size_t j = 1; static_cast<size_t>(1 << j) <= n; ++j) {
for (size_t i = 0; static_cast<size_t>(i + (1 << j)) <= n; ++i) {
/**
* @note notice how we deal with the range of length `pow(2,i)`,
* and we can reuse the computation that we did for the range of
* length `pow(2,i-1)`.
*
* So, ST[j][i] = min( ST[j-1][i], ST[j-1][i + pow(2,j-1)]).
* @example ST[2][3] = min(ST[1][3], ST[1][5])
*/
int64_t x = ST[j - 1][i]; ///< represents minimum value over
///< the range [j,i]
int64_t y =
ST[j - 1]
[i + (1 << (j - 1))]; ///< represents minimum value over
///< the range [j,i + pow(2,j-1)]
ST[j][i] =
(A[x] <= A[y] ? x : y); ///< represents minimum value over
///< the range [j,i]
}
}
}
/**
* @brief Queries the sparse table for the value of the interval [l, r]
* (i.e. from l to r inclusive).
* @param l the left index of the range (inclusive).
* @param r the right index of the range (inclusive).
* @return the computed value of the given interval.
* @complexity: O(1)
*/
int64_t query(int64_t l, int64_t r) {
int64_t g = LOG[r - l + 1]; ///< smallest power of 2 covering [l,r]
int64_t x = ST[g][l]; ///< represents minimum value over the range
///< [g,l]
int64_t y =
ST[g][r - (1 << g) + 1]; ///< represents minimum value over the
///< range [g, r - pow(2,g) + 1]
return (A[x] <= A[y] ? x : y); ///< represents minimum value over
///< the whole range [l,r]
}
};
} // namespace sparse_table
} // namespace data_structures
/**
* @brief Self-test implementations
* @returns void
*/
static void test() {
/* We take an array as an input on which we need to perform the ranged
* minimum queries[RMQ](https://en.wikipedia.org/wiki/Range_minimum_query).
*/
std::array<int64_t, 10> testcase = {
1, 2, 3, 4, 5,
6, 7, 8, 9, 10}; ///< array on which RMQ will be performed.
size_t testcase_size =
sizeof(testcase) / sizeof(testcase[0]); ///< size of self test's array
data_structures::sparse_table::Sparse_table
st{}; ///< declaring sparse tree
std::copy(std::begin(testcase), std::end(testcase),
std::begin(st.A)); ///< copying array to the struct
st.n = testcase_size; ///< passing the array's size to the struct
st.buildST(); ///< precomputing sparse tree
// pass queries of the form: [l,r]
assert(st.query(1, 9) == 1); ///< as 1 is smallest from 1..9
assert(st.query(2, 6) == 2); ///< as 2 is smallest from 2..6
assert(st.query(3, 8) == 3); ///< as 3 is smallest from 3..8
std::cout << "Self-test implementations passed!" << std::endl;
}
/**
* @brief Main function
* @param argc commandline argument count (ignored)
* @param argv commandline array of arguments (ignored)
* @returns 0 on exit
*/
int main(int argc, char *argv[]) {
test(); // run self-test implementations
return 0;
}