看python standard library by exmple里面提到一个Counter容器,它像muliset一样,能够维持一个集合,并在常量时间插入元素、查询某个元素的个数,而且还提供了一个
most_common(n)方法,用于统计频数最大的n个元素,这在读取文本并统计词频的时候显得非常实用。
考虑C++实现的时候,查到一个叫做LFU的东西,https://en.wikipedia.org/wiki/Least_frequently_used,是关于磁盘缓存策略的,基本想法跟这个counter有类似的地方。
http://dhruvbird.com/lfu.pdf 这里有相关的实现。
#include<iostream>
#include<list>
#include<vector>
#include<unordered_map>
using namespace std;
//关键字节点
template<typename T>
struct keyNode{
typedef T value_type;
keyNode(){}
keyNode(T v, keyNode* p, keyNode* n) :val(v), prev(p), next(n){}
T val;
keyNode* prev;
keyNode* next;
};
//计数器节点
template<typename T>
struct countNode{
countNode(){
keyhead = new keyNode<T> ;
keyhead->prev = keyhead->next = NULL;
}
~countNode(){
while (keyhead->next != NULL){
keyNode<T>* p = keyhead->next;
keyhead->next = p->next;
delete p;
}
delete keyhead;
}
countNode(int f, countNode* p, countNode *n):
freq(f),prev(p),next(n){
keyhead = new keyNode<T>;
keyhead->prev = keyhead->next = NULL;
}
keyNode<T>* insertKey(const T& v){
keyNode<T>* node = new keyNode<T>(v, keyhead, keyhead->next);
if (keyhead->next != NULL)
keyhead->next->prev = node;
keyhead->next = node;
return node;
}
int freq;
keyNode<T>* keyhead;
countNode* prev;
countNode* next;
};
//计数器容器
/***支持如下操作:
插入(insert) 时间复杂度O(1)
查找(lookup) 时间复杂度O(1)
查询最频繁的n个元素(most_common(n)) 时间复杂度o(n)
删除操作 时间复杂度o(1)
**/
template<typename T>
class Counter{
public:
Counter(){
head = new countNode<T>(0, NULL, NULL);
tail = NULL;
}
~Counter(){
while (head->next != NULL){
countNode<T>* p = head->next;
head->next = p->next;
delete p;
}
delete head;
}
//插入一个关键字,如果已经存在,频数加1
void insert(const T& v){
if (dict.find(v) == dict.end()){
//关键字是新插入的
if (head->next == NULL || head->next->freq != 1){
//需要新建count节点
countNode<T>* node = new countNode<T>(1, head, head->next);
if (head->next == NULL)
tail = node;
head->next = node;
dict[v] = pair<countNode<T>*, keyNode<T>*>(node, node->insertKey(v));
}
else{
dict[v] =
pair<countNode<T>*, keyNode<T>*>(head->next, head->next->insertKey(v));
}
}
else{
//关键字已经存在了
//频数必然会有增加,这时对结构的改动较大
countNode<T>* countAddr = dict[v].first;
countNode<T>* nextCount = countAddr->next;
keyNode<T>* keyAddr = dict[v].second;
int freq = countAddr->freq;
//首先从countAddr删除一个keyAddr节点
keyAddr->prev->next = keyAddr->next;
if (keyAddr->next != NULL)
keyAddr->next->prev = keyAddr->prev;
delete keyAddr;
if (nextCount == NULL || nextCount->freq != freq + 1){
//需要加一个countNode节点
countNode<T>* node = new countNode<T>(freq + 1, countAddr, nextCount);
if (nextCount != NULL)
nextCount->prev = node;
else
tail = node;
countAddr->next = node;
dict[v] =
pair<countNode<T>*, keyNode<T>*>(node, node->insertKey(v));
}
else{
dict[v] =
pair<countNode<T>*, keyNode<T>*>(nextCount, nextCount->insertKey(v));
}
//如果删除的keyNode节点是countNode中最后一个keyNode,就要把countAddr也删除了
if (countAddr->keyhead->next == NULL){
countAddr->prev->next = countAddr->next;
if (countAddr->next != NULL)
countAddr->next->prev = countAddr->prev;
delete countAddr;
}
}
}
//返回关键字的频数
int lookup(const T& v)const{
return dict[v].first->freq;
}
/**返回频数最高的n个元素
返回形式为:(key,count)
**/
vector<pair<T, int>> most_common(int n){
//链表的顺序是频数从低到高的,此时需要从尾节点逆向遍历n个元素
vector<pair<T, int>> result;
countNode<T>* countVisitor = tail;
while (n > 0 && countVisitor != NULL){
keyNode<T>* keyVisitor = countVisitor->keyhead->next;
while (n > 0 && keyVisitor != NULL){
result.emplace_back(keyVisitor->val, countVisitor->freq);
n--;
keyVisitor = keyVisitor->next;
}
countVisitor = countVisitor->prev;
}
return result;
}
vector<pair<T, int>> least_common(int n){
vector<pair<T, int>> result;
countNode<T>* countVisitor = head->next;
while (n > 0 && countVisitor != NULL){
keyNode<T>* keyVisitor = countVisitor->keyhead->next;
while (n > 0 && keyVisitor != NULL){
result.emplace_back(keyVisitor->val, countVisitor->freq);
n--;
keyVisitor = keyVisitor->next;
}
countVisitor = countVisitor->next;
}
return result;
}
private:
countNode<T>* head;
countNode<T>* tail;
unordered_map<T, pair<countNode<T>*, keyNode<T>*>> dict;
};
int main(){
{
Counter<char> wordCount;
string s("jfoaedfrerlkmgvj9ejajiokl;fdaks");
for (auto v : s){
wordCount.insert(v);
}
auto result = wordCount.least_common(3);
}
return 0;
}