views:

1385

answers:

5

I am building a plugin for a LAN party website that I wrote that would allow the use of a Round Robin tournament.

All is going well, but I have some questions about the most efficient way to rank over two criteria.

Basically, I would like the following rakning layout:

         Rank  Wins  TotalScore
PersonE  1     5     50
PersonD  2     3.5   37
PersonA  2     3.5   37
PersonC  4     2.5   26
PersonB  5     2.5   24
PersonF  6     0     12

In SQL server, I would use:

SELECT
    [Person],
    RANK() OVER (ORDER BY Wins DESC, TotalScore DESC) [Rank],
    [Wins],
    [TotalScore]

Now, I only have List, Dictionary, and etc. to work with

Specifically:

Dictionary<TournamentTeam, double> wins = new Dictionary<TournamentTeam, double>();
Dictionary<TournamentTeam, double> score = new Dictionary<TournamentTeam, double>();

Is there a way to do this style of ranking with LINQ?

If not, is there an extensible way that would allow me later to take in to account Win-Loss-Draw instead of just wins if I choose to?

Edit:

My adaptation of TheSoftwareJedi's answer:

private class RRWinRecord : IComparable
{
    public int Wins { get; set; }
    public int Losses { get; set; }
    public int Draws { get; set; }
    public double OverallScore { get; set; }
    public double WinRecord
    {
        get
        {
            return this.Wins * 1.0 + this.Draws * 0.5 + this.Losses * 0.0;
        }
    }

    public int CompareTo(object obj) { ... }

    public override bool Equals(object obj) { ... }
    public override int GetHashCode() { ... }
    public static bool operator ==(RRWinRecord lhs, RRWinRecord rhs) { ... }
    public static bool operator !=(RRWinRecord lhs, RRWinRecord rhs) { ... }
    public static bool operator >(RRWinRecord lhs, RRWinRecord rhs) { ... }
    public static bool operator <(RRWinRecord lhs, RRWinRecord rhs) { ... }
    public static bool operator >=(RRWinRecord lhs, RRWinRecord rhs) { ... }
    public static bool operator <=(RRWinRecord lhs, RRWinRecord rhs) { ... }
}

...

    int r = 1, lastRank = 1;
    RRWinRecord lastRecord = null;

    var ranks = from team in records.Keys
                let teamRecord = records[team]
                orderby teamRecord descending
                select new RRRank() { Team = team, Rank = r++, Record = teamRecord };

    foreach (var rank in ranks)
    {
        if (rank.Record != null && lastRecord == rank.Record)
        {
            rank.Rank = lastRank;
        }

        lastRecord = rank.Record;
        lastRank = rank.Rank;

        string scoreDescription = String.Format("{0}-{1}-{2}", rank.Record.Wins, rank.Record.Losses, rank.Record.Draws);
        yield return new TournamentRanking(rank.Team, rank.Rank, scoreDescription);
    }

    yield break;
+1  A: 

This could be a start:

Dictionary<TournamentTeam, double> wins = new Dictionary<TournamentTeam, double>();
Dictionary<TournamentTeam, double> score = new Dictionary<TournamentTeam, double>();
Dictionary<TournamentTeam, int> ranks = new Dictionary<TournamentTeam, int>();

int r = 1;

ranks = (
    from name 
    in wins.Keys 
    orderby wins[name] descending, scores[name] descending
    select new { Name = name, Rank = r++ })
    .ToDictionary(item => item.Name, item => item.Rank);
Philippe Leybaert
It's a great start. I'll see if I can't add the tie situation
TheSoftwareJedi
Yeah, great start +1. However, I need non-dense ranking. I building a NUnit test to target the desired behavior. Will check back soon.
John Gietzen
Added the non-dense solution below. Was going to just edit this, but I think it's so different it warranted a new answer.
TheSoftwareJedi
+2  A: 

Assuming you have a List<Result> structure where the Result object has the following parameters...

Pesron     - string
Rank       - int
Wins       - double
TotalScore - int

You could write a custom comparer, and then pass that to List.Sort(Comparison<Result> comparison)

Alternative, you could just make your Result object implement IComparable<Result> and stick this in your class.

        #region IComparable Members

        public int CompareTo(Result obj)
        {
            if (this.Rank.CompareTo(obj.Rank) != 0)
                return this.Rank.CompareTo(obj.Rank);

            if (this.Wins.CompareTo(obj.Wins) != 0)
                return (this.Wins.CompareTo(obj.Wins);

            return (this.TotalScore.CompareTo(obj.TotalScore) ;

        }

        #endregion

Then you can just call List<Result>.Sort();

Eoin Campbell
Don't get why this is voted down. You can simply implement your own comparer, with your own formula to determine the equality of two objects. Any formula will do!
Erik van Brakel
thanks eric. was wondering the same myself
Eoin Campbell
I did not vote this down, but I need non-dense ranking, which is unavailable in your solution.
John Gietzen
+1: Ah, now that I have cleaned up, I have added a class similar to yours, and I now see where this is useful. See the question for how I have ended up using all of the suggestions.
John Gietzen
+1  A: 

This should work for a non-dense rank:

static class Program
{

    static IEnumerable<Result> GetResults(Dictionary<TournamentTeam, double> wins, Dictionary<TournamentTeam, double> scores)
    {
        int r = 1;
        double lastWin = -1;
        double lastScore = -1;
        int lastRank = 1;

        foreach (var rank in from name in wins.Keys
                             let score = scores[name]
                             let win = wins[name]
                             orderby win descending, score descending
                             select new Result { Name = name, Rank = r++, Score = score, Win = win })
        {
            if (lastWin == rank.Win && lastScore == rank.Score)
            {
                rank.Rank = lastRank;
            }
            lastWin = rank.Win;
            lastScore = rank.Score;
            lastRank = rank.Rank;
            yield return rank;
        }
    }
}

class Result
{
    public TournamentTeam Name;
    public int Rank;
    public double Score;
    public double Win;
}
TheSoftwareJedi
+1 Almost Perfect! The one think I wish would be that the logic for ordering parameters was in a single place (rather than in the order by clause, and in the if statement.)
John Gietzen
Possible bug fix: set "lastRank" to 1 so that in case the best score somehow turns out to be "-1", the rank will still be set to 1.
John Gietzen
the logic for ordering is in one place. the logic for comparing is in another. you could put it into the Result class using CompareTo and Equals
TheSoftwareJedi
@John Gietzen - kudos. fix applied.
TheSoftwareJedi
+1  A: 

I realize I'm late to the party, but I wanted to take a shot anyhow.

Here is a version which uses LINQ exclusively:

private IEnumerable<TeamRank> GetRankings(Dictionary<TournamentTeam, double> wins, Dictionary<TournamentTeam, double> scores)
{
    var overallRank = 1;

    return
        from team in wins.Keys
        group team by new { Wins = wins[team], TotalScore = scores[team] } into rankGroup
        orderby rankGroup.Key.Wins descending, rankGroup.Key.TotalScore descending
        let currentRank = overallRank++
        from team in rankGroup
        select new TeamRank(team, currentRank, rankGroup.Key.Wins, rankGroup.Key.TotalScore);
}

The return type:

public class TeamRank
{
    public TeamRank(TournamentTeam team, int rank, double wins, double totalScore)
    {
        this.Team = team;
        this.Rank = rank;
        this.Wins = wins;
        this.TotalScore = totalScore;
    }

    public TournamentTeam Team { get; private set; }

    public int Rank { get; private set; }

    public double Wins { get; private set; }

    public double TotalScore { get; private set; }
}
Bryan Watts
I like this too.
John Gietzen
+2  A: 

Ranking isn't too hard. Just mishmash OrderBy and Select implementation patterns together and you can have an easy to use Ranking extension method. Like this:

    public static IEnumerable<U> Rank<T, TKey, U>
    (
      this IEnumerable<T> source,
      Func<T, TKey> keySelector,
      Func<T, int, U> selector
    )
    {
        if (!source.Any())
        {
            yield break;
        }

        int itemCount = 0;
        T[] ordered = source.OrderBy(keySelector).ToArray();
        TKey previous = keySelector(ordered[0]);
        int rank = 1;
        foreach (T t in ordered)
        {
            itemCount += 1;
            TKey current = keySelector(t);
            if (!current.Equals(previous))
            {
                rank = itemCount;
            }
            yield return selector(t, rank);
            previous = current;
        }
    }

Here's some test code

string[] myNames = new string[]
{ "Bob", "Mark", "John", "Jim", "Lisa", "Dave" };
//
var query = myNames.Rank(s => s.Length, (s, r) => new { s, r });
//
foreach (var x in query)
{
  Console.WriteLine("{0} {1}", x.r, x.s);
}

Which yields these results:

1 Bob
1 Jim
3 Mark
3 John
3 Lisa
3 Dave
David B
Nice! All I have to do is implement the proper operators on my class, etc. and I can rank by anything! woo!
John Gietzen
It's always good to get a "woo!" :)
David B