0

I need to execute some code on all the values of a key, and I need to repeated that action for every key. I look for something like:

for(auto key_iterator = hash_multimap.begin_keys();
    key_iterator != hash_multimap.end_keys(); key_iterator++)
{
    auto key = key_iterator->key;
    // set up state
    for(auto value_iterator = key_iterator->begin_values();
        value_iterator != key_iterator->end_values(); value_iterator++)
    {
        // mutate state
    }
    // use state
    // tear down state
}

This of course doesn't work, but is there a way to achieve similar effect? the problem is that I need to go over each key and then use a shared state for all of them. An example of what it can be used for is like:

typedef std::hash_multimap<int> hash_t;
typedef hash_t::value_type hash_val;

hash_t hash;
hash.insert(hash_val(0, 1));
hash.insert(hash_val(1, 2));
hash.insert(hash_val(1, 3));
hash.insert(hash_val(2, 4));
hash.insert(hash_val(2, 5));
hash.insert(hash_val(2, 6));
hash.insert(hash_val(3, 7));
hash.insert(hash_val(3, 8));
hash.insert(hash_val(3, 9));

// print out the sum of values for each key here.
// expected output:
//
// 0: 1
// 1: 5
// 2: 15
// 3: 24

The problem with just using hash_multimap.begin() is that I can't be sure it returns each key in consecutive block of that key, and even if it did, I can't know where such block begins and where it ends.

Edit: I also can't use hash_multimap.equal_range(key) because I can't iterate over the keys. a way to iterate over keys that includes every key only once will solve this too.

How can I do this?

Daniel
  • 30,896
  • 18
  • 85
  • 139

2 Answers2

1

You can use lower bound and upper bound to achieve this. For example try -

 auto value_iterator = hash->begin();
 while( value_iterator != hash->end() ){
 {
     auto lIter = hash->lower_bound( value_iterator->first );
     auto uIter = hash->upper_bound( value_iterator->first );
     while( lIter != uIter ){
         // sum the values associated with keys
         // Increment lIter
     }
     value_iterator = uIter;
 }

Edit: You can use equal_range as @Thanatos suggested if the library you are using doesn't have upper_bound, lower_bound member functions. Microsoft implementation does actually have these. The logic is same and the inner loop would be -

pair<hash_multimap<int,int>::iterator,hash_multimap<int,int>::iterator> pairIter;
pairIter.equal_range(value_Iterator->first);
while( pairIter.first != pairIter.second ){
    // sum the values associated with keys
    // Increment pairIter->first
}
value_iterator = pairIter.second;
Mahesh
  • 34,573
  • 20
  • 89
  • 115
0

Since you used C++11 syntax, I will as well ;-]

#include <algorithm>
#include <numeric>
#include <unordered_map>
#include <iostream>

int main()
{
    typedef std::unordered_multimap<int, int> map_t;
    typedef std::unordered_map<int, int> sums_t;

    map_t hash;
    hash.insert(map_t::value_type(0, 1));
    hash.insert(map_t::value_type(1, 2));
    hash.insert(map_t::value_type(1, 3));
    hash.insert(map_t::value_type(2, 4));
    hash.insert(map_t::value_type(2, 5));
    hash.insert(map_t::value_type(2, 6));
    hash.insert(map_t::value_type(3, 7));
    hash.insert(map_t::value_type(3, 8));
    hash.insert(map_t::value_type(3, 9));

    sums_t const& sums = std::accumulate(hash.cbegin(), hash.cend(), sums_t(),
        [](sums_t& acc, map_t::const_reference p)
        {
            return acc[p.first] += p.second, acc;
        }
    );

    std::for_each(sums.cbegin(), sums.cend(),
        [](sums_t::const_reference p)
        {
            std::cout << p.first << ": " << p.second << '\n';
        }
    );
}
ildjarn
  • 62,044
  • 9
  • 127
  • 211
  • Creating another map to store sum is a bit costly, considering that `hash_multimap` already sorts them in buckets internally. a side note: you should use `std::endl` instead of `'\n'`. – Daniel Sep 10 '11 at 04:08
  • 2
    @Dani : 1) It's not costly if you need to retain the values of your original map, which is likely most of the time. Additionally, the other solution has `O(4N•log2(N))` complexity, whereas this one has `O(N)` complexity. 2) No, you shouldn't; `std::endl` is merely shorthand for `'\n' << std::flush`, and I don't want the stream to be flushed for every single value (the C++ standard already guarantees that all standard iostreams will be flushed automatically upon program termination). – ildjarn Sep 10 '11 at 04:17