private LearningSet SplitOnTrainAndTest(InputsOutputsData inputOutputsData)
{
var allSamplesCount = inputOutputsData.Count;
var trainSamplesCount = GetTrainSampleCount(allSamplesCount);
var rand = new Random();
var learningSet = new LearningSet();
var trainingData = learningSet.TrainingData;
var testData = learningSet.TestData;
for (int i = 0, trainSamplesLeft = trainSamplesCount, samplesLeft = allSamplesCount;
i < allSamplesCount;
++i, --samplesLeft)
{
bool shouldTakeNextTrainSample = rand.Next(1, samplesLeft) <= trainSamplesLeft;
if (shouldTakeNextTrainSample)
{
trainingData.AddData(inputOutputsData.Inputs[i], inputOutputsData.Outputs[i]);
--trainSamplesLeft;
}
else
{
testData.AddData(inputOutputsData.Inputs[i], inputOutputsData.Outputs[i]);
}
}
return(learningSet);
}
private LearningSet GetCategoryLearningSet(Category category, int numberOfCategories)
{
if (!category.Files.Any())
{
var errorInfo = $"No files found in '{category.Name}' category.";
throw new InvalidOperationException(errorInfo);
}
var inputOutputsData = new InputsOutputsData();
var imagePreprocessingStrategy = _dataProviderconfiguration.ToImagePreprocessingStrategy();
foreach (var file in category.Files)
{
using (var image = (Bitmap)Image.FromFile(file, true))
using (var processedImage = imagePreprocessingStrategy.Process(image))
{
var input = ConvertImage(processedImage);
inputOutputsData.Inputs.Add(input);
var output = new double[numberOfCategories];
output[category.Index] = 1;
inputOutputsData.Outputs.Add(output);
}
}
var splittedSets = SplitOnTrainAndTest(inputOutputsData);
return(splittedSets);
}
public LearningSet()
{
TrainingData = new InputsOutputsData();
TestData = new InputsOutputsData();
}
