Hello,
Im using Collections.sort() to sort a LinkedList whose elements implements Comparable interface, so they are sorted in a natural order. In the javadoc documentation its said this method uses mergesort algorithm wich has n*log(n) performance.
My question is if there is a more efficient algorithm to sort my LinkedList?
The size of that list could be very high and sort will be also very frequent.
Thanks!
There is no general sort algorithm better than n*log(n). And this is quite fast. By general I mean your data doesn't have special properties.
In terms of sorting the list, no, all comparison based sorts on general data are O(N log(N)).
If your resorting is due to insertions, then you can try to batch your insertions and then merge sort with the main list - if you have B new items, you sort them in O(B log(B)) then do a single level merge of the two lists which is O(N+B).
If your resorting is due to changes in the values of the items, you might be able to do a similar batching if you change the mutable values into immutable ones and treat the changes to be a batch of insertions and deletions. Otherwise, you won't be able to avoid sorting the whole list.
If your requirements allow it, then there are various non-linked-list structures such as TreeSet available which maintain a sorted order more efficiently, but will fail if the values are mutable.
O(N log N) is very good asymptotically. That said, there are linear time O(N) non-comparison based sort, e.g. counting sort and bucket sort. This is useful when, e.g. you're sorting millions and millions of integers, but they're between 1..10.
Also, if the list is "almost sorted", the otherwise quadratic insertion sort is reported to actually be better under some scenarios.
Whether or not this is applicable, or even worth to implement, depends on your profiling results. I'd say that unless it shows the sort to be a bottleneck, don't worry about it.
If you say the list will be sorted "very frequent", you should consider holding the list in a sorted stated all the time, like using a tree instead of a LinkedList. Maybe you can even use some SortedSet instead of a List, if you don't have any duplicated values and don't need any List operations (as you are sorting them anyway all the time). Check the TreeSet class of the SortedSet implementation.
This implementation provides guaranteed log(n) time cost for the basic operations (add, remove and contains).
If you want to iterate over this "list" (which is actually a Set) you can use the Iterator of the class.
Returns an iterator over the elements in this set in ascending order.
If you have duplicate values inside the List you have to use some tricks (like putting the value in a new class which also got some delta for sorting equal object)