public void testsigmoidMatrix()
{
int hiddenlayer = 30;
string[] allLabel = { "1", "2", "3" };
int lengthImage = 10;
double learningRate = 0.1;
Matrix <double> transposePOutputWeightMatrix = DenseMatrix.Create(7, 1, 0);
for (int i = 0; i < 7; i++)
{
transposePOutputWeightMatrix[i, 0] = i + 1;
}
MNIST.CNN.CNN cnn = new MNIST.CNN.CNN(hiddenlayer, allLabel, lengthImage, learningRate);
Matrix <double> newOutput = cnn.sigmoidMatrix(transposePOutputWeightMatrix);
Matrix <double> expectMatrix = DenseMatrix.Create(7, 1, 0);
expectMatrix[0, 0] = 0.731058579;
expectMatrix[1, 0] = 0.880797078;
expectMatrix[2, 0] = 0.952574127;
expectMatrix[3, 0] = 0.98201379;
expectMatrix[4, 0] = 0.993307149;
expectMatrix[5, 0] = 0.997527377;
expectMatrix[6, 0] = 0.999088949;
Assert.AreEqual(true, equalMatrix(newOutput, expectMatrix, 0.0001));
}
public void testAnswer()
{
int hiddenlayer = 30;
string[] allLabel = { "0", "1", "2", "3", "4", "5", "6", "7", "8", "9" };
int lengthImage = 10;
double learningRate = 0.1;
MNIST.CNN.CNN cnn = new MNIST.CNN.CNN(hiddenlayer, allLabel, lengthImage, learningRate);
Matrix <double> outputPredic = DenseMatrix.Create(10, 1, 0);
outputPredic[0, 0] = 0.731058579;
outputPredic[1, 0] = 0.880797078;
outputPredic[2, 0] = 0.952574127;
outputPredic[3, 0] = 0.98201379;
outputPredic[4, 0] = 0.993307149;
outputPredic[5, 0] = 0.997527377;
outputPredic[6, 0] = 0.999088949;
outputPredic[7, 0] = 0.997527377;
outputPredic[8, 0] = 0.999999999;
outputPredic[9, 0] = 0.997527377;
string answer = cnn.answer(outputPredic);
string expect = "8";
Assert.AreEqual(answer, expect);
}
public void testSubtrackMatrix()
{
int hiddenlayer = 30;
string[] allLabel = { "0", "1", "2", "3", "4", "5", "6", "7", "8", "9" };
int lengthImage = 10;
double learningRate = 0.1;
MNIST.CNN.CNN cnn = new MNIST.CNN.CNN(hiddenlayer, allLabel, lengthImage, learningRate);
Matrix <double> matrix1 = DenseMatrix.Create(10, 1, 0);
matrix1[0, 0] = 10;
matrix1[1, 0] = 9;
matrix1[2, 0] = 8;
matrix1[3, 0] = 7;
matrix1[4, 0] = 6;
matrix1[5, 0] = 5;
matrix1[6, 0] = 4;
matrix1[7, 0] = 3;
matrix1[8, 0] = 2;
matrix1[9, 0] = 1;
Matrix <double> matrix2 = DenseMatrix.Create(10, 1, 0);
matrix2[0, 0] = 1;
matrix2[1, 0] = 2;
matrix2[2, 0] = 3;
matrix2[3, 0] = 4;
matrix2[4, 0] = 5;
matrix2[5, 0] = 6;
matrix2[6, 0] = 7;
matrix2[7, 0] = 8;
matrix2[8, 0] = 9;
matrix2[9, 0] = 10;
Matrix <double> expect = DenseMatrix.Create(10, 1, 0);
expect[0, 0] = 9;
expect[1, 0] = 7;
expect[2, 0] = 5;
expect[3, 0] = 3;
expect[4, 0] = 1;
expect[5, 0] = -1;
expect[6, 0] = -3;
expect[7, 0] = -5;
expect[8, 0] = -7;
expect[9, 0] = -9;
Matrix <double> subtractMatrix = matrix1.Subtract(matrix2);
Assert.AreEqual(subtractMatrix, expect);
}
public void testTransposeAndMultiply()
{
int hiddenlayer = 30;
string[] allLabel = { "1", "2", "3" };
int lengthImage = 10;
double learningRate = 0.1;
Matrix <double> matrix1 = DenseMatrix.Create(7, 1, 0);
for (int i = 0; i < 7; i++)
{
matrix1[i, 0] = i + 1;
}
Matrix <double> matrix2 = DenseMatrix.Create(3, 1, 0);
matrix2[0, 0] = 140;
matrix2[1, 0] = 336;
matrix2[2, 0] = 532;
MNIST.CNN.CNN cnn = new MNIST.CNN.CNN(hiddenlayer, allLabel, lengthImage, learningRate);
Matrix <double> newOutputMatrix = matrix2.TransposeAndMultiply(matrix1);
Matrix <double> exprect = DenseMatrix.Create(3, 7, 0);
int count = 0;
exprect[0, 0] = 140;
exprect[0, 1] = 280;
exprect[0, 2] = 420;
exprect[0, 3] = 560;
exprect[0, 4] = 700;
exprect[0, 5] = 840;
exprect[0, 6] = 980;
exprect[1, 0] = 336;
exprect[1, 1] = 672;
exprect[1, 2] = 1008;
exprect[1, 3] = 1344;
exprect[1, 4] = 1680;
exprect[1, 5] = 2016;
exprect[1, 6] = 2352;
exprect[2, 0] = 532;
exprect[2, 1] = 1064;
exprect[2, 2] = 1596;
exprect[2, 3] = 2128;
exprect[2, 4] = 2660;
exprect[2, 5] = 3192;
exprect[2, 6] = 3724;
Assert.AreEqual(newOutputMatrix, exprect);
}
public void testDerivativeSigmoid()
{
int hiddenlayer = 30;
string[] allLabel = { "0", "1", "2", "3", "4", "5", "6", "7", "8", "9" };
int lengthImage = 10;
double learningRate = 0.1;
MNIST.CNN.CNN cnn = new MNIST.CNN.CNN(hiddenlayer, allLabel, lengthImage, learningRate);
Matrix <double> matrix1 = DenseMatrix.Create(10, 1, 0);
matrix1[0, 0] = 0.904083039728828;
matrix1[1, 0] = 0.00000941094029599077;
matrix1[2, 0] = 0.0106577785865162;
matrix1[3, 0] = 0.00703778917745922;
matrix1[4, 0] = 0.000820703909257055;
matrix1[5, 0] = 0.00243879162789633;
matrix1[6, 0] = 0.0120445706347156;
matrix1[7, 0] = 0.000936247465062854;
matrix1[8, 0] = 0.00992988293783782;
matrix1[9, 0] = 0.0000295245568786306;
Matrix <double> expect = DenseMatrix.Create(10, 1, 0);
expect[0, 0] = 0.08671689700351050000;
expect[1, 0] = 0.00000941085173019352;
expect[2, 0] = 0.010544190342117;
expect[3, 0] = 0.00698825870095285;
expect[4, 0] = 0.000820030354350385;
expect[5, 0] = 0.00243284392329204;
expect[6, 0] = 0.011899498952941;
expect[7, 0] = 0.000935370905747017;
expect[8, 0] = 0.00983128036267865;
expect[9, 0] = 0.0000295236851791717;
Matrix <double> derivaticesSigmoidMatrix = cnn.derivativeSigmoid(matrix1);
Assert.AreEqual(equalMatrix(derivaticesSigmoidMatrix, expect, 0.0000000001), true);
}
public void testAddMatrix()
{
int hiddenlayer = 30;
string[] allLabel = { "1", "2", "3" };
int lengthImage = 10;
double learningRate = 0.1;
Matrix <double> pDeltasMatrix = DenseMatrix.Create(7, 1, 0);
for (int i = 0; i < 7; i++)
{
pDeltasMatrix[i, 0] = i + 1;
}
Matrix <double> transposePOutputWeightMatrix = DenseMatrix.Create(7, 1, 0);
for (int i = 0; i < 7; i++)
{
transposePOutputWeightMatrix[i, 0] = i + 1;
}
MNIST.CNN.CNN cnn = new MNIST.CNN.CNN(hiddenlayer, allLabel, lengthImage, learningRate);
Matrix <double> newOutputMatrix = transposePOutputWeightMatrix.Add(pDeltasMatrix);
Matrix <double> expectMatrix = DenseMatrix.Create(7, 1, 0);
expectMatrix[0, 0] = 2;
expectMatrix[1, 0] = 4;
expectMatrix[2, 0] = 6;
expectMatrix[3, 0] = 8;
expectMatrix[4, 0] = 10;
expectMatrix[5, 0] = 12;
expectMatrix[6, 0] = 14;
Assert.AreEqual(newOutputMatrix, expectMatrix);
}
public void testConsFunction()
{
int hiddenlayer = 30;
string[] allLabel = { "1", "2", "3" };
int lengthImage = 10;
double learningRate = 0.1;
double[] predic = new double[] { 1, 2, 3, 4, 5, 6, 7 };
double[] target = new double[] { 1, 2, 3, 5, 5, 5, 5 };
MNIST.CNN.CNN cnn = new MNIST.CNN.CNN(hiddenlayer, allLabel, lengthImage, learningRate);
double error;
double expect = 3;
if (cnn.conFunc(predic, target, out error))
{
Assert.AreEqual(error, expect);
}
else
{
Assert.AreEqual(0, 1);
}
}
public void TestMulMatrix()
{
int hiddenlayer = 30;
string[] allLabel = { "1", "2", "3" };
int lengthImage = 10;
double learningRate = 0.1;
Matrix <double> pDeltasMatrix = DenseMatrix.Create(7, 1, 0);
for (int i = 0; i < 7; i++)
{
pDeltasMatrix[i, 0] = i + 1;
}
Matrix <double> transposePOutputWeightMatrix = DenseMatrix.Create(3, 7, 0);
int count = 0;
for (int i = 0; i < 3; i++)
{
for (int j = 0; j < 7; j++)
{
transposePOutputWeightMatrix[i, j] = ++count;
}
}
MNIST.CNN.CNN cnn = new MNIST.CNN.CNN(hiddenlayer, allLabel, lengthImage, learningRate);
Matrix <double> newOutputMatrix = transposePOutputWeightMatrix.Multiply(pDeltasMatrix);
Matrix <double> expectMatrix = DenseMatrix.Create(3, 1, 0);
expectMatrix[0, 0] = 140;
expectMatrix[1, 0] = 336;
expectMatrix[2, 0] = 532;
Assert.AreEqual(newOutputMatrix, expectMatrix);
}