public TrainerSettings()
{
EpochsCount = 1;
Optimizer = new CpuAdam(1e-3f);
LossFunction = new CrossEntropy();
Metric = new CpuClassificationAccuracy();
}
public void CrossEntropyTest()
{
var crossEntropyOH = new CrossEntropyOneHot();
var crossEntropy = new CrossEntropy();
int nb = 10;
double[] input = new double[nb];
double[] output = new double[nb];
for (int i = 0; i < nb; i++)
{
input[i] = Math.PI * rnd.NextDouble();
}
var softmax = new Softmax();
softmax.Activate(input, output);
double[] expectedOutput = new double[nb];
expectedOutput[output.ArgMax()] = 1;
double[] errors = new double[nb];
double entropyOH = crossEntropyOH.Evaluate(output, expectedOutput, errors);
double entropy = crossEntropy.Evaluate(output, expectedOutput, errors);
Assert.AreEqual(entropyOH, entropy);
}
public void CrossEntropyTest()
{
var actual = new Matrix(4, 1);
var expected = new Matrix(4, 1);
actual.InRandomize();
expected.InRandomize();
var autoErr = new CrossEntropy().Evaluate(actual, expected);
var error = 0.0;
for (var i = 0; i < actual.Rows; i++)
{
for (var j = 0; j < actual.Columns; j++)
{
error += -expected[i, j] * Math.Log(actual[i, j]) - (1.0 - expected[i, j]) * Math.Log(1.0 - actual[i, j] + double.Epsilon);
}
}
error /= actual.Rows * actual.Columns;
Assert.IsTrue(Math.Abs(error - autoErr) < 0.01, new CrossEntropy().Type().ToString() + " Forward!");
var autoDErr = new CrossEntropy().Backward(actual, expected);
var dErr = (actual - expected).HadamardDivision(actual.Hadamard(actual) - (actual + actual.Fill(double.Epsilon)));
Assert.IsTrue(Math.Abs(dErr.FrobeniusNorm() - autoDErr.FrobeniusNorm()) < 0.01, new CrossEntropy().Type().ToString() + " Backward!");
}
public void AccuracyTest2()
{
var y = new TensorOld(new double[] { 0, 0, 1, 0, 1, 0 }, 2, 3);
var yHat = new TensorOld(new double[] { 0, 1, 0, 0, 1, 0 }, 2, 3);
var expected = 0.5;
var actual = CrossEntropy.ComputeAccuracy(y, yHat);
Assert.Equal(expected, actual);
}
public void LossStorageTest()
{
var loss = new CrossEntropy();
var xml = new XmlDocument();
var el = XmlStorage.SaveToEl(xml, loss);
var test = XmlStorage.LoadFromNode <CrossEntropy>(el);
Assert.True(test is CrossEntropy);
}
public void ForwardTest2()
{
var yHat = new TensorOld(new double[] { 0.05, 0.15, 0.7, 0.1 }, 1, 4);
var yHat2 = new TensorOld(new double[] { 0.1, 0.15, 0.65, 0.1 }, 1, 4);
var y = new TensorOld(new double[] { 0, 0, 1, 0 }, 1, 4);
var ce = new CrossEntropy();
ce.PrepareTrain(y, yHat);
ce.Compute(y, yHat);
var loss = ce.GetLoss();
ce.Compute(y, yHat2);
var loss2 = ce.GetLoss();
Assert.True(loss < loss2);
}
public void ForwardTest()
{
//模拟sigmoid的情况，4个样本
var yHat = new TensorOld(new double[] { 0.7, 0.2, 0.4, 0.9 }, 4, 1);
var yHat2 = new TensorOld(new double[] { 0.7, 0.2, 0.6, 0.9 }, 4, 1);
var y = new TensorOld(new double[] { 1, 0, 1, 1 }, 4, 1);
var ce = new CrossEntropy();
ce.PrepareTrain(y, yHat);
ce.Compute(y, y);
var minLoss = ce.GetLoss();
ce.Compute(y, yHat);
var loss = ce.GetLoss();
ce.Compute(y, yHat2);
var loss2 = ce.GetLoss();
Assert.Equal(0, minLoss);
Assert.True(loss > minLoss);
Assert.True(loss > loss2);
}
