Revenge of kNN II
Time Limit: 8000/5000 MS (Java/Others) Memory Limit: 32768/32768 K (Java/Others)Total Submission(s): 292 Accepted Submission(s): 91
Problem Description
In pattern recognition, the k-Nearest Neighbors algorithm (or k-NN for short) is a non-parametric method used for classification and regression. In both cases, the input consists of the k closest training examples in the feature space.
In k-NN regression, the output is the property value for the object. This value is the average of the values of its k nearest neighbors.
---Wikipedia
Today, kNN takes revenge on you, again. You have to handle a kNN case in one-dimensional coordinate system. There are N points with a position Xi and value Vi. Then there are M kNN queries for point with index i, recalculate its value by averaging the values its k-Nearest Neighbors. Note you have to replace the value of i-th point with the new calculated value. And if there is a tie while choosing k-Nearest Neighbor, choose the one with the minimal index first.
(Have you ever tried the problem “Revenge of kNN”? They are twin problems!)
In k-NN regression, the output is the property value for the object. This value is the average of the values of its k nearest neighbors.
---Wikipedia
Today, kNN takes revenge on you, again. You have to handle a kNN case in one-dimensional coordinate system. There are N points with a position Xi and value Vi. Then there are M kNN queries for point with index i, recalculate its value by averaging the values its k-Nearest Neighbors. Note you have to replace the value of i-th point with the new calculated value. And if there is a tie while choosing k-Nearest Neighbor, choose the one with the minimal index first.
(Have you ever tried the problem “Revenge of kNN”? They are twin problems!)
Input
The first line contains a single integer T, indicating the number of test cases.
Each test case begins with two integers N and M. Then N lines follows, each line contains two integers Xi and Vi. Then M lines with the queried index Qi and Ki follows, in which Ki indicating the number of k-Nearest Neighbors
[Technical Specification]
1. 1 <= T <= 5
2. 2 <= N <= 100 000
3. 1 <= M <= 100 000
4. 1 <= Vi <= 1 000
5. 1 <= Xi <= 1 000 000 000, and no two Xi are identical.
6. 1 <= Qi <= N
7. 1 <= Ki <= N - 1
Each test case begins with two integers N and M. Then N lines follows, each line contains two integers Xi and Vi. Then M lines with the queried index Qi and Ki follows, in which Ki indicating the number of k-Nearest Neighbors
[Technical Specification]
1. 1 <= T <= 5
2. 2 <= N <= 100 000
3. 1 <= M <= 100 000
4. 1 <= Vi <= 1 000
5. 1 <= Xi <= 1 000 000 000, and no two Xi are identical.
6. 1 <= Qi <= N
7. 1 <= Ki <= N - 1
Output
For each test case, output sum of all queries rounded to three fractional digits.
Sample Input
1 5 3 1 2 2 3 3 6 4 8 5 8 2 2 3 2 4 2
Sample Output
17.000HintFor the first query, the 2-NN for point 2 is point 1 and 3, so the new value is (2 + 6) / 2 = 4. For the second query, the 2-NN for point 3 is point 2 and 4, and the value of point 2 is changed to 4 by the last query, so the new value is (4 + 8) / 2 = 6. Huge input, faster I/O method is recommended.
思路:用二分取查找半径,然后树状数组求和即可。
代码:
#include <iostream>
#include <cstdio>
#include <cstring>
#include <algorithm>
#include <cmath>
using namespace std;
#define N 100050
int dis[N];
double c[N];
struct Node
{
int x,id;
double v;
bool operator < (const Node &a)const{
return x<a.x;
}
} p[N];
bool cmp(Node a,Node b)
{
return a.x<b.x;
}
int finds_high(int x,int n)
{
if(x>=p[n].x) return n;
int l=1,r=n,ans;
while(l<=r)
{
int mid=(l+r)>>1;
if(p[mid].x>x) r=mid-1;
else ans=mid,l=mid+1;
}
return ans;
}
int finds_down(int x,int n)
{
if(x<=p[1].x) return 1;
int l=1,r=n,ans;
while(l<=r)
{
int mid=(l+r)>>1;
if(p[mid].x>=x) ans=mid,r=mid-1;
else l=mid+1;
}
return ans;
}
int lowbit(int x)
{
return x&-x;
}
void add(int x,double v,int n)
{
for(int i=x; i<=n; i+=lowbit(i))
c[i]+=v;
}
double sum(int x)
{
double ans=0;
for(int i=x; i; i-=lowbit(i))
ans+=c[i];
return ans;
}
int main()
{
int T,n,m;
int q,k;
scanf("%d",&T);
while(T--)
{
scanf("%d %d",&n,&m);
for(int i=0; i<=100000; i++)
c[i]=0;
for(int i=1; i<=n; i++)
{
scanf("%d %lf",&p[i].x,&p[i].v);
p[i].id=i;
}
sort(p+1,p+n+1,cmp);
for(int i=1; i<=n; i++)
{
dis[p[i].id]=i;
add(i,p[i].v,n);
}
double ans=0;
while(m--)
{
scanf("%d %d",&q,&k);
int id=dis[q];
int d=p[id].x;
int l=0,r=1000000000;
double s=0;
if(k>=n) ans=ans+(sum(n)-p[id].v)/(n-1);
else
{
while(l<=r)
{
int mid=(l+r)>>1;
/*int rr=finds_high(d+mid,n);
int ll=finds_down(d-mid,n);*/
Node t;
t.x=d+mid;
int rr=upper_bound(p+1,p+1+n,t)-p-1;
t.x=d-mid;
int ll=lower_bound(p+1,p+1+n,t)-p;
if(rr-ll==k+1&&p[rr].x-p[id].x==p[id].x-p[ll].x)
{
if(p[rr].id<p[ll].id) ll++;
else rr--;
s=s+sum(rr)-sum(ll-1)-p[id].v;
break;
}
else if(rr-ll==k)
{
s=s+sum(rr)-sum(ll-1)-p[id].v;
break;
}
else if(rr-ll<k) l=mid+1;
else if(rr-ll>k) r=mid-1;
}
add(id,s/k-p[id].v,n);
p[id].v=s/k;
ans+=p[id].v;
}
}
printf("%.3lf\n",ans);
}
return 0;
}