public IEnumerable<double> RunLearner(DataSet dataSet, Learner learner, StreamWriter outputWriter, StreamWriter csvWriter, int iteration)
{
double[] runScores = new double[numberOfRuns];
double[] runTimes = new double[numberOfRuns];
double[] runTicks = new double[numberOfRuns];
Stopwatch watch = new Stopwatch();
for (int i = 0; i < numberOfRuns; i++)
{
Console.WriteLine("Run {0}...", (i + 1));
watch.Reset();
dataSet.SplitData(2.0 / 3.0, i);
watch.Start();
learner.Learn(dataSet.TrainingData, dataSet.ValidationData, dataSet.TestData);
watch.Stop();
double score = 0.0;
if (useTestData)
{
score = PautomacEvaluator.Evaluate(learner, dataSet.TestData, dataSet.SolutionData);
}
else
{
foreach (int[] signal in dataSet.ValidationData.GetAll())
{
score -= learner.CalculateProbability(signal, true);
}
}
runScores[i] = score;
runTimes[i] = (watch.ElapsedMilliseconds / 1000.0);
runTicks[i] = watch.ElapsedTicks;
outputWriter.WriteLine("Run {0:00}:\t{1:00000000.0000000000}\t{2:000000}\t{3:0000000000000000}", (i + 1), runScores[i], runTimes[i], runTicks[i]);
outputWriter.Flush();
csvWriter.WriteLine("{0},{1},{2},{3},{4},", iteration, i, runScores[i], runTimes[i], runTicks[i]);
csvWriter.Flush();
using (StreamWriter modelWriter = new StreamWriter(String.Format(@"Benchmark_{0}/DataSet_{1}/Models_{2}/Iter{3}_Run{4}.txt", Name, dataSet.Number, learner.Name().ToLowerInvariant().Replace(' ', '_'), iteration, i)),
modelCSVWriter = new StreamWriter(String.Format(@"Benchmark_{0}/DataSet_{1}/Models_{2}/Iter{3}_Run{4}.csv", Name, dataSet.Number, learner.Name().ToLowerInvariant().Replace(' ', '_'), iteration, i)))
{
modelWriter.AutoFlush = true;
modelCSVWriter.AutoFlush = true;
modelWriter.WriteLine("DataSet {0}", dataSet.Number);
modelWriter.WriteLine("Learner: {0}", learner.Name());
modelWriter.WriteLine("{0}: {1:0000.0000000000}", (useTestData ? "PautomaC Score" : "Log Likelihood"), score);
modelWriter.WriteLine();
learner.Save(modelWriter, modelCSVWriter);
}
}
yield return runScores.Average();
yield return Median(runScores);
yield return runTimes.Average();
yield return Median(runTimes);
}
public IEnumerable<double> BenchmarkLearner(DataSet dataSet, Learner learner)
{
Console.WriteLine("Benchmarking Learner {0}...", learner.Name());
//This sets up a file for intermediate output during learning
if (learner is Learners.GreedyExtendLearner) {
string intermediateOutputFileName = String.Format(@"Benchmark_{0}/DataSet_{1}/{2}", Name, dataSet.Number, "intermediate");
((Learners.GreedyExtendLearner)learner).SetIntermediateOutputFile(intermediateOutputFileName);
((Learners.GreedyExtendLearner)learner).SetSolutions(dataSet.SolutionData);
}
if (learner is Learners.PadawanLearner) {
string intermediateOutputFileName = String.Format(@"Benchmark_{0}/DataSet_{1}/{2}", Name, dataSet.Number, "intermediate");
((Learners.PadawanLearner)learner).SetIntermediateOutputFile(intermediateOutputFileName);
}
if (learner is Learners.GGLearner)
{
string intermediateOutputFileName = String.Format(@"Benchmark_{0}/DataSet_{1}/{2}", Name, dataSet.Number, "intermediate");
((Learners.GGLearner)learner).SetIntermediateOutputFile(intermediateOutputFileName);
}
Dictionary<int, double> parameterAverageScores = new Dictionary<int, double>();
Dictionary<int, double> parameterMedianScores = new Dictionary<int, double>();
Dictionary<int, double> parameterAverageRuntimes = new Dictionary<int, double>();
Dictionary<int, double> parameterMedianRuntimes = new Dictionary<int, double>();
int bestIteration = 0;
LearnerParameters parameters = learners[learner];
using (StreamWriter outputWriter = new StreamWriter(String.Format(@"Benchmark_{0}/DataSet_{1}/{2}.txt", Name, dataSet.Number, learner.Name().ToLowerInvariant().Replace(' ', '_'))),
csvSummaryWriter = new StreamWriter(String.Format(@"Benchmark_{0}/DataSet_{1}/{2}_SUMMARY.csv", Name, dataSet.Number, learner.Name().ToLowerInvariant().Replace(' ', '_'))),
csvResultWriter = new StreamWriter(String.Format(@"Benchmark_{0}/DataSet_{1}/{2}_RESULTS.csv", Name, dataSet.Number, learner.Name().ToLowerInvariant().Replace(' ', '_'))))
{
outputWriter.AutoFlush = true;
csvSummaryWriter.AutoFlush = true;
csvResultWriter.AutoFlush = true;
outputWriter.WriteLine("DataSet {0}", dataSet.Number);
outputWriter.WriteLine("Learner: {0}", learner.Name());
outputWriter.WriteLine();
csvSummaryWriter.WriteLine("Iteration,Median Score,Average Score,Median Time,Average Time,");
csvResultWriter.WriteLine("Iteration,Run,Score,Time,Ticks,");
for (int i = 0; ((parameters.Minimum + (i * parameters.StepSize)) <= parameters.Maximum); i++)
{
Console.WriteLine("Benchmarking Model with {0}...", IterationName(parameters, i));
outputWriter.WriteLine("{0}:", IterationName(parameters, i));
outputWriter.WriteLine();
learner.Initialise(parameters, i);
double[] results = RunLearner(dataSet, learner, outputWriter, csvResultWriter, i).ToArray();
parameterAverageScores.Add(i, results[0]);
parameterMedianScores.Add(i, results[1]);
parameterAverageRuntimes.Add(i, results[2]);
parameterMedianRuntimes.Add(i, results[3]);
if ((bestIteration < 0) || (parameterMedianScores[bestIteration] > parameterMedianScores[i]))
{
bestIteration = i;
}
csvSummaryWriter.WriteLine("{0},{1},{2},{3},{4}", i, parameterMedianScores[i], parameterAverageScores[i], parameterMedianRuntimes[i], parameterAverageRuntimes[i]);
csvSummaryWriter.Flush();
outputWriter.WriteLine();
if (parameters.StepSize == 0)
{
break;
}
}
outputWriter.WriteLine();
outputWriter.WriteLine("SUMMARY");
outputWriter.WriteLine();
foreach (int iteration in parameterMedianScores.Keys)
{
outputWriter.WriteLine("{0}:\t{1:00000000.0000000000}\t{2:00000000.0000000000}\t{3:000000}\t{4:000000}", IterationName(parameters, iteration), parameterMedianScores[iteration],
parameterAverageScores[iteration], parameterMedianRuntimes[iteration], parameterAverageRuntimes[iteration]);
}
outputWriter.WriteLine();
outputWriter.WriteLine("BEST");
outputWriter.WriteLine();
outputWriter.WriteLine("{0}:\t{1:00000000.0000000000}\t{2:00000000.0000000000}\t{3:000000}\t{4:000000}", IterationName(parameters, bestIteration), parameterMedianScores[bestIteration],
parameterAverageScores[bestIteration], parameterMedianRuntimes[bestIteration], parameterAverageRuntimes[bestIteration]);
}
yield return parameterAverageScores[bestIteration];
yield return parameterMedianScores[bestIteration];
yield return parameterAverageRuntimes[bestIteration];
yield return parameterMedianRuntimes[bestIteration];
}