static JobResult RunJob(JobDescription job)
{
if (job.HiddenNeurons == 0)
{
job.Ns = new int[] { 2, 1 }
}
;
else
{
job.Ns = new int[] { 2, job.HiddenNeurons, 1 }
};
Random rng = new Random();
NeuralNetwork nn = new NeuralNetwork(job.Beta, job.Ns);
nn.RandomizeWeights(-0.2, 0.2);
nn.RandomizeBiases(-1, 1);
int iters = 200000;
for (int iter = 0; iter < iters; iter++)
{
int iPattern = rng.Next(0, job.TrainingData.Length);
nn.Train(job.TrainingData[iPattern], job.TrainingClasses[iPattern], job.LearningRate);
}
if (Dump)
{
Task4a.WriteDump("task4b_dump_" + job.HiddenNeurons + "_" + DateStr, nn);
Interlocked.Increment(ref JobsCompleted);
return(null);
}
double trainingErrorRate = Task4a.ErrorRate(job.TrainingData, job.TrainingClasses, nn);
double validationErrorRate = Task4a.ErrorRate(job.ValidationData, job.ValidationClasses, nn);
Interlocked.Increment(ref JobsCompleted);
return(new JobResult
{
HiddenNeurons = job.HiddenNeurons,
Training = trainingErrorRate,
Validation = validationErrorRate
});
}
}
public static void Run(IEnumerable <string> args)
{
int nRuns;
if (!Task4a.ParseArgs("task4b", ref args, out nRuns, out Dump))
{
return;
}
DateStr = DateTime.Now.ToString("yyyy-MM-dd-HH-mm-ss");
Matrix <double>[][] dataAll;
int[][][] classesAll;
Task4a.ReadNormalizeSplit(out classesAll, out dataAll, args.First(), args.Skip(1).First());
int[][] trainingClasses = classesAll[0], validationClasses = classesAll[1];
Matrix <double>[] trainingData = dataAll[0], validationData = dataAll[1];
double Beta = 0.5;
double learningRate = 0.01;
int[] hiddenNeurons = { 0, 2, 4, 8, 16, 32 };
var jobs = hiddenNeurons.Select(hn => new JobDescription
{
HiddenNeurons = hn,
Beta = Beta,
LearningRate = learningRate,
TrainingData = trainingData,
TrainingClasses = trainingClasses,
ValidationData = validationData,
ValidationClasses = validationClasses
});
var runs = Enumerable.Repeat(jobs, nRuns).SelectMany(x => x);
TotalJobs = runs.Count();
Thread progress = new Thread(() => Progress.ProgressFunc(ref TotalJobs, ref JobsCompleted));
progress.Start();
var results = runs.AsParallel().Select(RunJob).ToList();
if (!Dump)
{
var byNeurons = results.GroupBy(res => res.HiddenNeurons).OrderBy(n => n.Key);
string filename = "task4b_" + nRuns + "_" + DateStr + ".txt";
Console.WriteLine("Writing to " + filename + "...");
using (StreamWriter sw = new StreamWriter(new FileStream(filename, FileMode.CreateNew), Encoding.ASCII))
{
foreach (var byNeuron in byNeurons)
{
double trainingErrorRate = byNeuron.Average(res => res.Training);
double validationErrorRate = byNeuron.Average(res => res.Validation);
sw.WriteLine(byNeuron.Key + " " +
trainingErrorRate.ToString(CultureInfo.InvariantCulture) + " " +
validationErrorRate.ToString(CultureInfo.InvariantCulture));
}
}
string errorLog = "task4b_" + nRuns + "_" + DateStr + ".log";
Console.WriteLine("Executing MATLAB script...");
if (!MATLAB.RunScript(errorLog, "Task4bGrapher", "'" + filename + "'"))
{
Console.WriteLine("An error occured while running MATLAB, check the log\n\tLog file:" + errorLog);
}
}
else
{
string errorRateFile = Directory.EnumerateFiles(Directory.GetCurrentDirectory(), "task4b_*_*.txt")
.Where(file => !file.StartsWith("task4b_dump_"))
.OrderBy(file => file)
.LastOrDefault();
if (errorRateFile == null)
{
Console.WriteLine("Unable to find an error rates file, skipping MATLAB script...");
}
else
{
Console.WriteLine(errorRateFile);
string errorLog = "task4b_dump_" + DateStr + ".log";
string dumpname = "task4b_dump_%d_" + DateStr + ".txt";
Console.WriteLine("Executing MATLAB script...");
if (!MATLAB.RunScript(errorLog, "Task4bExtra", "'" + dumpname + "'", "'" + errorRateFile + "'"))
{
Console.WriteLine("An error occured while running MATLAB, check the log\n\tLog file:" + errorLog);
}
}
}
Console.WriteLine("Done!");
}
