5

The scenario is a timeline of events, that I want to be able to query for all items within a specific date range.

I am looking for a data structure in .NET (up to v4.0) that stores items as sorted and unique (for example, by using a comparer or a unique key). It should support adding/deleting at no more than logarithmic complexity, and performing binary search at that complexity as well.

System.Collections.Generic.SortedSet seemed like what I wanted, but its GetViewBetween() method returns an inclusive list of items, as a SortedSet.

I am missing two things in it:

  1. Calling ToList() or enumerating the SortedSet is too expensive, as the list is long. I need the method return a List<T>, not a SortedSet<T>.
  2. Calling it with an inclusive/exclusive date range, to my choice - not possible and I need that too.

If you know a good library that contains such a data structure, that is tested and familiar, I would sure like to try it.

Thanks.

Dor Rotman
  • 1,511
  • 1
  • 15
  • 27
  • Have you looked at the SortedList(Of TKey, TValue) class? – Chris Gessler Mar 11 '12 at 18:12
  • 1
    did you check SortedDictionary? – Asdfg Mar 11 '12 at 18:12
  • 1
    You should be able to use LINQ with `SortedSet` to do everything you're looking for. – M.Babcock Mar 11 '12 at 18:14
  • Are items added in date order? – tvanfosson Mar 11 '12 at 18:19
  • @M.Babcock: LINQ uses "foreach" inside as far as I know. This works in O(n) complexity for querying the date range. – Dor Rotman Mar 11 '12 at 22:13
  • @ChrisGessler: I did check it out. I need to iterate over the Keys collection in order to find the beginning of the range I want to query. It is O(n) complexity. I can find a piece of code that performs binary search over this collection, but I saw that this subject is complex and would rather have a data structure that is more meant for my needs. – Dor Rotman Mar 11 '12 at 22:19
  • @Asdfg: I did check it out. Please see above answer, it's the same issue. – Dor Rotman Mar 11 '12 at 22:29
  • @tvanfosson: No, items are added according to user input in no particular order. – Dor Rotman Mar 11 '12 at 22:29
  • Just to be clear, what is the desired time boundary for the `GetViewBetween` method as you want it (Not the `SortedSet` one)? – ntziolis Mar 12 '12 at 00:43
  • @ntziolis: It should find the beginning of the range in logarithmic complexity, but then the enumerating can be O(n) since the range may be the whole input. – Dor Rotman Mar 12 '12 at 08:50

2 Answers2

1

After a bit of reading, it would seem that SortedList or SortedDictionary is what you need. SortedList appears to use less memory and is technically a binary tree, where SortedDictionary is faster with unsorted data. Beyond that, they are very close cousins.

Here's a good question/answer on the differences: SortedList vs. SortedDictionary vs. Sort()

Community
  • 1
  • 1
Chris Gessler
  • 22,727
  • 7
  • 57
  • 83
  • Both of them lack the ability to retrieve a beginning of a range of items in logarithmic complexity. For instance, my keys are dates with time set to midnight, and I need a range that starts after some day at noon. I have to search the Keys list for the value I need, and would need to implement a binary search to do this in logarithmic complexity. – Dor Rotman Mar 11 '12 at 22:31
1

The problem as I see it is that you want a data structure organized on two axes - your unique key and by time. If you can trade space for time, I'd suggest different data structures (wrapped in your own class to ensure consistency) for each. You might want to use a SortedList to keep track of your key-oriented data. I believe it's based on a Red-Black Tree and should have the characteristics you desire for key-search. Alternatively, if you don't need them ordered by key, you could use a simple Dictionary.

To support date-based search, you might want to have a B-tree (one implementation, note I haven't tested it: http://blog.daisley-harrison.com/blog/post/NET-Generic-BTree-Library-and-Source-Code.aspx) keyed by date. Make sure that it supports duplicate keys, however, since these might not be unique. It can contain either a copy of the data or simply the key associated with that timestamp.

All of these structures have log(n), or better, complexity for search. Listing of items between two dates should be pretty efficient, with the best performance coming with the B-Tree/Dictionary combination.

tvanfosson
  • 524,688
  • 99
  • 697
  • 795
  • I totally agree, there are two problems here that can only be solved within the specified time boundaries by sacrificing space – ntziolis Mar 12 '12 at 00:41
  • Actually, I need the information organized on one axis - DateTime (unique). Meaning, only a single event can happen at a given time. – Dor Rotman Mar 13 '12 at 10:25
  • Sorry, I understood it as needing time and unique-key based. A self-balancing tree should meet your requirements. SortedDictionary uses a Red-Black tree, but you also need it to expose the tree traversal directly since you won't be doing lookup but "nearest" in your search. AFAIK, there aren't any tree implementations that expose the traversal in the framework. You might want to check codeplex or Google for a .NET tree data structure. Here's one I found: http://www.itu.dk/research/c5/ – tvanfosson Mar 13 '12 at 12:46