public void DataTransformers()
{
PakiraDecisionTreeGenerator pakiraGenerator = PakiraGeneratorTests.CreatePakiraGeneratorInstance();
TrainData trainData = new TrainData();
trainData = trainData.AddSample(ImmutableList.CreateRange(new double[] { 2, 3 }), 42);
trainData = trainData.AddSample(ImmutableList.CreateRange(new double[] { 120, 140 }), 54);
trainData = trainData.AddSample(ImmutableList.CreateRange(new double[] { 190, 200 }), 42);
PassThroughTransformer passThroughTransformer = new PassThroughTransformer();
MeanDistanceDataTransformer meanDistanceDataTransformer = new MeanDistanceDataTransformer();
Converter <IEnumerable <double>, IEnumerable <double> > dataTransformers = null;
dataTransformers += passThroughTransformer.ConvertAll;
dataTransformers += meanDistanceDataTransformer.ConvertAll;
pakiraGenerator.CertaintyScore = 10.0;
PakiraDecisionTreeModel pakiraDecisionTreeModel = new PakiraDecisionTreeModel(dataTransformers, trainData.Samples[0]);
pakiraDecisionTreeModel = pakiraGenerator.Generate(pakiraDecisionTreeModel, trainData);
pakiraDecisionTreeModel.Tree.Root.ShouldNotBeNull();
pakiraDecisionTreeModel.PredictNode(trainData.Samples[0]).Value.ShouldBe(trainData.Labels[0]);
pakiraDecisionTreeModel.PredictNode(trainData.Samples[1]).Value.ShouldBe(trainData.Labels[1]);
pakiraDecisionTreeModel.PredictNode(trainData.Samples[2]).Value.ShouldBe(trainData.Labels[2]);
// The data transformers should allow to produce a very shallow tree
pakiraDecisionTreeModel.Tree.GetNodes().Count().ShouldBe(3);
}
public void CertaintyScore()
{
PakiraDecisionTreeGenerator pakiraGenerator = PakiraGeneratorTests.CreatePakiraGeneratorInstance();
pakiraGenerator.CertaintyScore = 1.0;
TrainData trainData = new TrainData();
trainData = trainData.AddSample(ImmutableList.CreateRange(new double[] { 2, 90 }), 42);
trainData = trainData.AddSample(ImmutableList.CreateRange(new double[] { 250, 140 }), 54);
trainData = trainData.AddSample(ImmutableList.CreateRange(new double[] { 200, 100 }), 42);
PakiraDecisionTreeModel pakiraDecisionTreeModel = new PakiraDecisionTreeModel(trainData.Samples[0]);
pakiraDecisionTreeModel = pakiraGenerator.Generate(pakiraDecisionTreeModel, trainData);
pakiraDecisionTreeModel.Tree.Root.ShouldNotBeNull();
pakiraDecisionTreeModel.PredictNode(trainData.Samples[0]).Value.ShouldBe(trainData.Labels[0]);
pakiraDecisionTreeModel.PredictNode(trainData.Samples[1]).Value.ShouldBe(trainData.Labels[1]);
pakiraDecisionTreeModel.PredictNode(trainData.Samples[2]).Value.ShouldBe(trainData.Labels[2]);
pakiraDecisionTreeModel.PredictNode(new SabotenCache(trainData.Samples[0])).PakiraLeaf.Value.ShouldBe(trainData.Labels[0]);
pakiraDecisionTreeModel.PredictNode(new SabotenCache(trainData.Samples[1])).PakiraLeaf.Value.ShouldBe(trainData.Labels[1]);
pakiraDecisionTreeModel.PredictNode(new SabotenCache(trainData.Samples[2])).PakiraLeaf.Value.ShouldBe(trainData.Labels[2]);
}
public void DeepTree()
{
PakiraDecisionTreeGenerator pakiraGenerator = PakiraGeneratorTests.CreatePakiraGeneratorInstance();
TrainData trainData = new TrainData();
trainData = trainData.AddSample(ImmutableList.CreateRange(new double[] { 2, 3 }), 42);
trainData = trainData.AddSample(ImmutableList.CreateRange(new double[] { 250, 254 }), 54);
trainData = trainData.AddSample(ImmutableList.CreateRange(new double[] { 250, 255 }), 42);
trainData = trainData.AddSample(ImmutableList.CreateRange(new double[] { 251, 253 }), 6);
trainData = trainData.AddSample(ImmutableList.CreateRange(new double[] { 251, 254 }), 9);
trainData = trainData.AddSample(ImmutableList.CreateRange(new double[] { 1, 2 }), 96);
trainData = trainData.AddSample(ImmutableList.CreateRange(new double[] { 2, 1 }), 97);
trainData = trainData.AddSample(ImmutableList.CreateRange(new double[] { 2, 2 }), 98);
trainData = trainData.AddSample(ImmutableList.CreateRange(new double[] { 3, 2 }), 99);
PakiraDecisionTreeModel pakiraDecisionTreeModel = new PakiraDecisionTreeModel(trainData.Samples[0]);
pakiraGenerator.MinimumSampleCount = 100;
pakiraDecisionTreeModel = pakiraGenerator.Generate(pakiraDecisionTreeModel, trainData);
pakiraDecisionTreeModel.Tree.Root.ShouldNotBeNull();
pakiraDecisionTreeModel.Tree.GetNodes().Count().ShouldBeGreaterThanOrEqualTo(15, "If the test fails because of this, the number can be reduced as long as it stays 'high'. Instead, the tree depth could also be validated.");
pakiraDecisionTreeModel.PredictNode(trainData.Samples[0]).Value.ShouldBe(trainData.Labels[0]);
pakiraDecisionTreeModel.PredictNode(trainData.Samples[1]).Value.ShouldBe(trainData.Labels[1]);
pakiraDecisionTreeModel.PredictNode(trainData.Samples[2]).Value.ShouldBe(trainData.Labels[2]);
}
public void MinimumSampleCount()
{
PakiraDecisionTreeGenerator pakiraGenerator = PakiraGeneratorTests.CreatePakiraGeneratorInstance();
TrainData trainData = new TrainData();
trainData = trainData.AddSample(ImmutableList.CreateRange(new double[] { 2, 3 }), 42);
trainData = trainData.AddSample(ImmutableList.CreateRange(new double[] { 20, 140 }), 54);
trainData = trainData.AddSample(ImmutableList.CreateRange(new double[] { 33, 200 }), 42);
PakiraDecisionTreeModel pakiraDecisionTreeModel = new PakiraDecisionTreeModel(trainData.Samples[0]);
pakiraDecisionTreeModel = pakiraGenerator.Generate(pakiraDecisionTreeModel, trainData);
pakiraDecisionTreeModel.Tree.Root.ShouldNotBeNull();
pakiraDecisionTreeModel.PredictNode(trainData.Samples[0]).Value.ShouldBe(trainData.Labels[0]);
pakiraDecisionTreeModel.PredictNode(trainData.Samples[1]).Value.ShouldBe(trainData.Labels[1]);
pakiraDecisionTreeModel.PredictNode(trainData.Samples[2]).Value.ShouldBe(trainData.Labels[2]);
}
static void Main(string[] args)
{
const int featureWindowSize = /*24*/ 16;
const int halfFeatureWindowSize = featureWindowSize / 2;
string imageMainPath = args[0];
Image <L8> fullTextImage = Image.Load <L8>(imageMainPath + @"Images\FS18800114.2.11-a2-427w-c32.png");
Image <L8> completeA = Image.Load <L8>(imageMainPath + @"\Images\FS18800114.2.11-a2-427w-c32\a\CompleteA.png");
PakiraDecisionTreeGenerator pakiraGenerator = new PakiraDecisionTreeGenerator();
TrainData trainData = new TrainData();
L8 whitePixel = new L8(255);
L8 blackPixel = new L8(0);
L8 dontCarePixel = new L8(128);
byte dontCarePixelValue = dontCarePixel.PackedValue;
byte[] imageCropPixelsData = new byte[featureWindowSize * featureWindowSize * Unsafe.SizeOf <L8>()];
completeA.ProcessPixelRows(accessor =>
{
Span <byte> imageCropPixels = new Span <byte>(imageCropPixelsData);
for (int y = 0; y < accessor.Height; y++)
{
Span <L8> pixelRow = accessor.GetRowSpan(y);
for (int x = 0; x < pixelRow.Length; x++)
{
// Get a reference to the pixel at position x
ref L8 pixel = ref pixelRow[x];
if (pixel != dontCarePixel)
{
Image <L8> whiteWindow = new Image <L8>(featureWindowSize, featureWindowSize, whitePixel);
Image <L8> imageCrop = whiteWindow.Clone(clone => clone.DrawImage(fullTextImage, new Point(halfFeatureWindowSize - x, halfFeatureWindowSize - y), 1));
imageCrop.CopyPixelDataTo(imageCropPixels);
trainData = trainData.AddSample(imageCropPixelsData.Select <byte, double>(s => s), pixel.PackedValue);
}
}
}
});