/// <summary>
/// The Success Rate is the percentage of the independent runs which reached specified accuracy levels.
/// A total of numLevels accuracy levels have been considered, with the accuracy levels starting at the best
/// accuracy obtained by all algorithms, logarithmically increasing towards the worst accuracy.
/// </summary>
/// <param name="algorithms">Algorithms.</param>
/// <param name="numLevels">The number of accuracy levels.</param>
public static void AddSuccessRate(List <Algorithm> algorithms, int numLevels)
{
double bestFit = double.MaxValue;
double worstFit = double.MinValue;
var fitnesses = algorithms.Select(x => x.GetFitness());
foreach (Measurement measure in fitnesses)
{
worstFit = Math.Max(worstFit, measure.FinalIteration.Max);
bestFit = Math.Min(bestFit, measure.FinalIteration.Min);
}
List <double> accuracyLevels = LogarithmicallyIncreasingScale(bestFit, worstFit, numLevels);
foreach (Algorithm algorithm in algorithms)
{
Measurement measure = algorithm.GetFitness();
double runs = measure.FinalIteration.Values.Count;
List <double> successRate = new List <double>(numLevels);
//TODO: this can be improved majorly!
foreach (double level in accuracyLevels)
{
//TODO: (1 - success) as R expects minimization and success rate is maximization
successRate.Add(1 - measure.FinalIteration.Values.Count(x => x < level) / runs);
}
Measurement result = new Measurement("SuccessRate");
IterationStats stats = new IterationStats(1, successRate);
result.AddIterationStatistics(1, stats);
algorithm.AddMeasurement(result);
}
}
/// <summary>
/// Adds the iteration statistics to the list.
/// </summary>
/// <param name="iteration">The iteration which we are adding.</param>
/// <param name="stats">The iteration statistics to add.</param>
public void AddIterationStatistics(int iteration, IterationStats stats)
{
_iterationStatistics.TryAdd(iteration, stats);
if (iteration > maxIters)
{
maxIters = iteration; //check if the new iteration is greater than previous maximum
}
}
/// <summary>
/// The Consistency is the sum of the squared differences between the fitness and the average fitnesses.
/// </summary>
/// <param name="fitness">Fitness.</param>
public static Measurement Consistency(Measurement fitness)
{
double avg = fitness.FinalIteration.Average;
Measurement result = new Measurement("Consistency");
IterationStats stats = new IterationStats(1, fitness.FinalIteration.Values.Select(x => Math.Pow(x - avg, 2)));
result.AddIterationStatistics(1, stats);
return(result);
}
protected virtual void ReadLine(string line)
{
string[] tokens = line.Split(splitChars, StringSplitOptions.RemoveEmptyEntries);
int iter;
if (!int.TryParse(tokens[0], out iter))
{
Console.WriteLine("Iteration is not the first column.");
}
string columnName;
int index = 1; //the index of the column being parsed
//loop through each column
while (index < columnArray.Length)
{
columnName = columnArray[index];
List <double> values = new List <double>();
//loop through the various columns (i.e., runs) for this measure
while (index <= columnArray.Length && columnArray[index - 1] == columnName)
{
values.Add(double.Parse(tokens[index]));
index++;
}
IterationStats stats = new IterationStats(iter, values);
measurements.AddOrUpdate(columnName,
x => //add function
{
Measurement meas = new Measurement(columnName);
meas.AddIterationStatistics(iter, stats);
return(meas);
},
(name, meas) => //update function
{
meas.AddIterationStatistics(iter, stats);
return(meas);
}
);
}
}
