C++, fast remove elements from vector unique to another vector - c++

Question

There are 2 unsorted vectors of int and vector of pairs int, int
std::vector <int> v1;
std::vector <std::pair<int, float> > v2;
containing millions of items.
How to remove as fast as possible such items from v1, that are unique to v2.first (ie not included in v2.first)?
Example:
v1: 5 3 2 4 7 8
v2: {2,8} {7,10} {5,0} {8,9}
----------------------------
v1: 3 4

Answer 1

There are two tricks I would use to do this as quickly as possible:
Use some sort of associative container (probably std::unordered_set) to store all of the integers in the second vector to make it dramatically more efficient to look up whether some integer in the first vector should be removed.
Optimize the way in which you delete elements from the initial vector.
More concretely, I'd do the following. Begin by creating a std::unordered_set and adding all of the integers that are the first integer in the pair from the second vector. This gives (expected) O(1) lookup time to check whether or not a specific int exists in the set.
Now that you've done that, use the std::remove_if algorithm to delete everything from the original vector that exists in the hash table. You can use a lambda to do this:
std::unordered_set<int> toRemove = /* ... */
v1.erase(std::remove_if(v1.begin(), v1.end(), [&toRemove] (int x) -> bool {
return toRemove.find(x) != toRemove.end();
}, v1.end());
This first step of storing everything in the unordered_set takes expected O(n) time. The second step does a total of expected O(n) work by bunching all the deletes up to the end and making lookups take small time. This gives a total of expected O(n)-time, O(n) space for the entire process.
If you are allowed to sort the second vector (the pairs), then you could alternatively do this in O(n log n) worst-case time, O(log n) worst-case space by sorting the vector by the key, then using std::binary_search to check whether a particular int from the first vector should be eliminated or not. Each binary search takes O(log n) time, so the total time required is O(n log n) for the sorting, O(log n) time per element in the first vector (for a total of O(n log n)), and O(n) time for the deletion, giving a total of O(n log n).
Hope this helps!

Answer 2

Assuming that neither container is sorted and that sorting is actually too expensive or memory is scarce:
v1.erase(std::remove_if(v1.begin(), v1.end(),
[&v2](int i) {
return std::find_if(v2.begin(), v2.end(),
[](const std::pair<int, float>& p) {
return p.first == i; })
!= v2.end() }), v1.end());
Alternatively sort v2 on first and use a binary search instead. If there is enough memory use an unordered_set to sort the first of v2.
Complete C++03 version:
#include <iostream>
#include <vector>
#include <utility>
#include <algorithm>
struct find_func {
find_func(int i) : i(i) {}
int i;
bool operator()(const std::pair<int, float>& p) {
return p.first == i;
}
};
struct remove_func {
remove_func(std::vector< std::pair<int, float> >* v2)
: v2(v2) {}
std::vector< std::pair<int, float> >* v2;
bool operator()(int i) {
return std::find_if(v2->begin(), v2->end(), find_func(i)) != v2->end();
}
};
int main()
{
// c++11 here
std::vector<int> v1 = {5, 3, 2, 4, 7, 8};
std::vector< std::pair<int, float> > v2 = {{2,8}, {7,10}, {5,0}, {8,9}};
v1.erase(std::remove_if(v1.begin(), v1.end(), remove_func(&v2)), v1.end());
// and here
for(auto x : v1) {
std::cout << x << std::endl;
}
return 0;
}