Question

My code processes a huge number of values and I’m looking for an efficient structure to keep track of the top (N) values, where N is less than 10, so collecting ALL numbers then sorting the list and taking the first (N) is probably not the most efficient way.

To do that, I’m building a collection of fixed size N, to keep the top (N) values sorted in descending order. The Add(T value) method of the sorted collection would add the value to the collection if value is higher than any of the existing values (in which case the last element is removed) or if the collection is not full.

I was able to implement what I wanted using a doubly LinkedList<T> since it has fast insertion and removal, but I was wondering if using SortedDictionary<TKey, TValue> or a priority queue would be better ?

Thank you.

Answer 1

The performance may really change.

For N < 10 any overly complex data structure will likely drag performance significantly (though perhaps not catastrophically) so I’d use an array to store the items.

Then there are 3 main possibilities on how to arrange the items in the array:

1. sorted is probably the best choice to keep things simple:
   • constant time to determine whether to insert a new item (compare with lowest)
   • O(N) time to insert – but this only happens for items that are in the N best-so-far. And if your input is sufficiently random, the average time will be even lower because most insertions will only move a few elements at the bottom of the top.
2. unsorted:
   • O(N) time for each input element, that’s too much compared to “sorted”
3. binary heap that implements a priority queue: more complex to implement but maybe even faster than “sorted”
   • constant time to determine whether to insert a new item (compare with lowest)
   • O(log N) time to insert – and this only happens for items that are in the N best-so-far