CodeForces - 626E 三分

题意：给n个数，从这n个数里，任意选一些数，使得这些数的平均值减去中位数尽可能的大

精度卡的好恶心啊。。。。。

首先发现集合个数一定是奇数，任何一个个数为偶数的集合，都可以通过去掉中间较大的那个数，来增大或不变这个值

如1 2 3 4 可以去掉3变成1 2 4

排序后，枚举每个数做中位数，然后三分集合个数

确定中位数之后，剩下的数肯定是越大越好，

会发现这个值并不是随着集合个数单调变化的，所以要三分求凸点。

直接double除法，会有精度问题，可以转换成乘法比较

换了好几种三分形式，都不对。。。。各种怀疑。。。。并不是很理解。。

ac代码：

#include<bits/stdc++.h>
using namespace std;
const int maxn = 2e5 + 10;
int a[maxn], n;
long long s[maxn];
long long fun(int x, int cnt){
	int l = x - cnt;
	int r = n - cnt + 1;
	long long sum = s[n] - s[r - 1] + s[x] - s[l - 1];
	return sum - 1ll * (2 * cnt + 1) * a[x];
}
int main(){
	cin >> n;
	for(int i = 1; i <= n; i++){
		scanf("%d", &a[i]);
	}
	sort(a + 1, a + n + 1);
	if(n <= 2){
		cout << 1 << endl;
		cout << a[1] << endl;
		return 0;
	}
	for(int i = 1; i <= n; i++){
		s[i] = s[i - 1] + a[i];
	}
	long long minn = 0, ans_index = 1, ans_count = 0;
	for(int i = 2; i < n; i++){
		int le = 1, ri = min(i - 1, n - i);
		long long minn_now = 0, ans = 0;
		while(le <= ri){
			int mid1 = (le + ri) >> 1;
			int mid2 = (ri + 1 + mid1) >> 1;
			long long f1 = fun(i, mid1);
			long long f2 = fun(i, mid2);
			if(f1 * (2 * mid2 + 1) < f2 * (2 * mid1 + 1)){
				le = mid1 + 1;
			}
			else{
				ri = mid2 - 1;
			}
		}
		ans = le, minn_now = fun(i, le); 
	//	cout << i << " " << ans << endl;
		if(minn_now * (2 * ans_count + 1) > minn * (2 * ans + 1)){
			ans_index = i;
			ans_count = ans;
			minn = minn_now;
		} 
	}
	cout << ans_count * 2 + 1 << endl;
	for(int i = ans_index - ans_count; i <= ans_index; i++)
	cout << a[i] << " ";
	for(int i = n - ans_count + 1; i <= n; i++)
	cout << a[i] << " ";
	cout << endl;
	return 0;
} 

wa的代码：

#include<bits/stdc++.h>
using namespace std;
const int maxn = 2e5 + 10;
int a[maxn], n;
long long s[maxn];
long long fun(int x, int cnt){
	int l = x - cnt;
	int r = n - cnt + 1;
	long long sum = s[n] - s[r - 1] + s[x] - s[l - 1];
	return sum - 1ll * (2 * cnt + 1) * a[x];
}
int main(){
	cin >> n;
	for(int i = 1; i <= n; i++){
		scanf("%d", &a[i]);
	}
	sort(a + 1, a + n + 1);
	if(n <= 2){
		cout << 1 << endl;
		cout << a[1] << endl;
		return 0;
	}
	for(int i = 1; i <= n; i++){
		s[i] = s[i - 1] + a[i];
	}
	long long minn = 0, ans_index = 1, ans_count = 0;
	for(int i = 2; i < n; i++){
		int le = 1, ri = min(i - 1, n - i);
		long long minn_now = 0, ans = 0;
		while(le < ri){
			int mid1 = (2 * le + ri) / 3;
			int mid2 = (le + 2 * ri) / 3;
			long long f1 = fun(i, mid1);
			long long f2 = fun(i, mid2);
			if(f1 * (2 * mid2 + 1) <= f2 * (2 * mid1 + 1)){
				le = mid1 + 1;
			}
			else{
				ri = mid2 - 1;
			}
		}
		ans = le, minn_now = fun(i, le); 
		if(minn_now * (2 * ans_count + 1) > minn * (2 * ans + 1)){
			ans_index = i;
			ans_count = ans;
			minn = minn_now;
		} 
	}
	cout << ans_count * 2 + 1 << endl;
	for(int i = ans_index - ans_count; i <= ans_index; i++)
	cout << a[i] << " ";
	for(int i = n - ans_count + 1; i <= n; i++)
	cout << a[i] << " ";
	cout << endl;
	return 0;
} 

也可以二分来做，比较一下mid+1和mid-1就可以了