static void Main(string[] args)
{
CultureInfo customCulture = (CultureInfo)System.Threading.Thread.CurrentThread.CurrentCulture.Clone();
customCulture.NumberFormat.NumberDecimalSeparator = ".";
System.Threading.Thread.CurrentThread.CurrentCulture = customCulture;
/*
* Tinker();
* Console.ReadLine();
* Environment.Exit(0);
*/
NetworkModel model = null;
bool teaching = false;
int epochRuns = 1;
int repeats = 1;
string inputFile = null;
CommandLine.Parser.Default.ParseArguments <Options>(args)
.WithParsed <Options>(opts =>
{
teaching = opts.Teaching;
epochRuns = opts.EpochRuns;
repeats = opts.Repeats;
Debug = opts.Debug;
#if !DEBUG
try
{
#endif
model = NetworkModel.Load(opts.Model);
inputFile = model.Path(opts.Input);
#if !DEBUG
Console.Clear();
}
catch (Exception e)
{
Console.WriteLine(e.Message);
Environment.Exit(1);
}
#endif
});
if (model is null)
{
Environment.Exit(0);
}
if (teaching)
{
Console.WriteLine("Launching model in teaching mode");
// Save newly generated model.
if (!model.Loaded)
{
model.Save();
}
var plotModel = new PlotModel {
Title = String.Format("{0} model \"{1}\" learning graph", model.Config.Type, model.Name)
};
var series = new OxyPlot.Series.LineSeries();
try
{
var graphMatrix = DelimitedReader.Read <double>(model.Path("graph"));
for (int i = 0; i < graphMatrix.RowCount; i++)
{
series.Points.Add(new DataPoint(i + 1, graphMatrix.Row(i).At(0)));
}
}
catch { }
plotModel.Series.Add(series);
plotModel.Axes.Add(new OxyPlot.Axes.LinearAxis
{
Minimum = 1,
Position = OxyPlot.Axes.AxisPosition.Bottom,
Title = "Epochs",
});
plotModel.Axes.Add(new OxyPlot.Axes.LinearAxis
{
Position = OxyPlot.Axes.AxisPosition.Left,
Title = "Cost",
});
var epochOffset = series.Points.Count;
var epoch = epochOffset;
var epochRunStep = epochRuns;
var stopwatch = new Stopwatch();
var epochStopwatch = new Stopwatch();
bool forceStop = false;
stopwatch.Restart();
var repeat = 1;
while (epochRuns == 0 || epoch < epochRuns + epochOffset)
{
Console.Write("\nepoch {0} / {1}...", ++epoch, epochOffset + epochRuns);
var previous = model.GetInfo();
epochStopwatch.Restart();
stopwatch.Start();
var cost = model.RunEpoch();
stopwatch.Stop();
Console.Write(" {0:f2} + {1:f2}s,", stopwatch.Elapsed.TotalSeconds, epochStopwatch.Elapsed.TotalSeconds);
Console.Write(" cost {0}", cost);
series.Points.Add(new DataPoint(epoch, cost));
if (model.Config.Threshold != null && Math.Abs(cost) <= model.Config.Threshold)
{
Console.WriteLine("\nError threshold ({0}) reached. Finished learning.", model.Config.Threshold);
forceStop = true;
}
if (epochRuns == 0 || epoch >= epochRuns + epochOffset || forceStop)
{
forceStop = false;
epochRuns += epochRunStep;
var pngExporter = new PngExporter();
pngExporter.ExportToFile(plotModel, model.Path("learning-graph.png"));
if (repeat++ >= repeats)
{
DisplayActions(model, series);
}
else
{
SaveModel(model, series);
}
}
}
}
else
{
if (model.DataTransformer is Model.Transformers.VectorDataTransformer)
{
Console.WriteLine("Running execution loop\n");
while (true)
{
Console.WriteLine("Model info:");
Console.WriteLine("===========");
Console.WriteLine(model.GetInfo());
DisplayActions(model);
var inputs = Vector <double> .Build.Dense(model.Config.Inputs);
for (int i = 0; i < model.Config.Inputs; i++)
{
inputs[i] = ReadDouble(String.Format("Input[{0}]", i));
}
Console.Clear();
Console.WriteLine("Inputs:");
Console.WriteLine("=======");
Console.WriteLine(inputs.ToVectorString());
Console.WriteLine("Result:");
Console.WriteLine("=======");
foreach (var label in model.Run(inputs))
{
Console.WriteLine(label);
}
Console.WriteLine();
}
}
else if (model.DataTransformer is Model.Transformers.MnistDataTransformer)
{
if (!File.Exists(inputFile))
{
Console.WriteLine("Input file '{0}' doesn't exist", inputFile);
Environment.Exit(1);
}
var inputs = model.InputTransformer.Transform(inputFile).Row(0);
foreach (var label in model.Run(inputs))
{
Console.WriteLine(label);
}
}
else
{
Console.WriteLine("Unsupported model input transformer '{0}'.", model.DataTransformer);
Environment.Exit(1);
}
}
}
