private static ITreeProgram <TOutput> GetLargestCommon(
ref ITreeProgram <TOutput> prog1, ref ITreeProgram <TOutput> prog2,
ISet <ITreeProgram <TOutput> > ignorePrograms = null)
{
// gets sub-programs and sort descendingly
var subProgs1 = new SortedSet <ITreeProgram <TOutput> >(prog1.GetSubPrograms(),
Comparer <ITreeProgram <TOutput> > .Create((a, b) => - (a.Length == b.Length
? a.CompareTo(b)
: a.Length.CompareTo(b.Length))));
var subProgs2 = new HashSet <ITreeProgram <TOutput> >(prog2.GetSubPrograms());
return(subProgs1.FirstOrDefault(
subProg1 =>
(ignorePrograms == null || !ignorePrograms.Contains(subProg1)) && subProgs2.Contains(subProg1)));
}
private static IDictionary <ITreeProgram <TOutput>, uint> GetSubPrograms(ITreeProgram <TOutput> program)
{
// organizes sub-programs in dictionary-count form
var subProgs = program.GetSubPrograms();
var subProgsCounts = new Dictionary <ITreeProgram <TOutput>, uint>();
foreach (var subProg in subProgs)
{
if (!subProgsCounts.ContainsKey(subProg))
{
subProgsCounts.Add(subProg, 0);
}
subProgsCounts[subProg]++;
}
return(subProgsCounts);
}
private static double Calculate(ITreeProgram <TOutput> prog1, ITreeProgram <TOutput> prog2)
{
// checks simple cases
if (prog1 == null || prog2 == null)
{
return(0);
}
if (prog1.Equals(prog2))
{
return(1);
}
// replace all common sub-programs by weighted variables
var ignorePrograms = new HashSet <ITreeProgram <TOutput> >();
var minCount = Math.Min(prog1.Length, prog2.Length);
for (var i = 0; i < minCount; i++)
{
var largestCommon = GetLargestCommon(ref prog1, ref prog2, ignorePrograms);
if (largestCommon == null || largestCommon.Length < 2)
{
break;
}
var newSubProgram = new WeightedVariable(largestCommon.Expression, largestCommon.Length);
prog1 = prog1.Replace(largestCommon, newSubProgram);
prog2 = prog2.Replace(largestCommon, newSubProgram);
ignorePrograms.Add(newSubProgram);
}
// gets sub-programs
var subProgs1 = new List <ITreeProgram <TOutput> > {
prog1
};
subProgs1.AddRange(prog1.GetSubPrograms());
var subProgs2 = new List <ITreeProgram <TOutput> > {
prog2
};
subProgs2.AddRange(prog2.GetSubPrograms());
// tries to align both trees in all possible ways
var minRelCost = double.MaxValue;
for (var i = 0u; i < prog1.Length; i++)
{
for (var j = 0u; j < prog2.Length; j++)
{
// alignment is determined by the starting indexes of both trees
var idx1 = i;
var idx2 = j;
// adds cost of having to add previous nodes (shifting/alignment cost)
var cost = i + j;
var totalCost = cost;
// gets edit cost between sub-programs
GetEditCost(subProgs1, subProgs2, ref idx1, ref idx2, ref cost, ref totalCost);
// checks cost of having to add remaining nodes
var remainder = prog1.Length - idx1 + prog2.Length - idx2 - 2;
cost += remainder;
totalCost += remainder;
var relCost = (double)cost / totalCost;
// updates min relative cost
if (relCost < minRelCost)
{
minRelCost = relCost;
}
}
}
return(1d - minRelCost);
}