public void ModelTrainingConverges()
{
ImageData img = ResourceManager.UsingDebugBitmap("testgrid1px.png", (bmp) => {
return ImageData.FromImage(bmp);
});
Label[,] labels = new Label[img.XSites,img.YSites];
Classification cfc = new Classification(labels);
ModelBuilder mbr = new ModelBuilder(ConcatenateFeatures.INSTANCE, LinearBasis.INSTANCE, 1d, 3000, 1d, 1d);
Model mfm = mbr.PseudoLikelihoodTrain("","", new ImageData[1]{img},new Classification[1]{cfc});
Assert.AreNotEqual(mfm.TimeToConverge, mbr.MaxIters);
}
public void CanClassify()
{
ImageData img = ResourceManager.UsingDebugBitmap("testgrid1px.png", (bmp) => {
return ImageData.FromImage(bmp);
});
Label[,] labels = new Label[img.XSites,img.YSites];
Classification cfc = new Classification(labels);
ModelBuilder mbr = new ModelBuilder(ConcatenateFeatures.INSTANCE, LinearBasis.INSTANCE, 1d, 3000, 1d, 1d);
Model mfm = mbr.PseudoLikelihoodTrain("", "", new ImageData[1]{img},new Classification[1]{cfc});
Classification inferred = mfm.MaximumAPosterioriInfer(img);
Assert.IsNotNull(inferred);
}
public Classification LogisticInfer(ImageData test_input)
{
Classification curr_classification = new Classification(new Label[test_input.XSites, test_input.YSites]);
for(int x = 0; x < test_input.XSites; x++) for(int y = 0; y < test_input.YSites; y++)
{
double modeled_prob_of_one = MathWrapper.Sigma(W.DotProduct(Transformer.Transform(test_input[x,y])));
//double prob_one = ((double)OnsSeen)/((double) SitesSeen);
//double prob_zero = 1d - prob_one;
double lambda = MathWrapper.Log(modeled_prob_of_one) - MathWrapper.Log (1 - modeled_prob_of_one)/* + MathWrapper.Log (prob_one/prob_zero)*/;
if(lambda > 0)
curr_classification[x,y] = Label.ON;
else
curr_classification[x,y] = Label.OFF;
}
return curr_classification;
}
