private Matrix createRandomMatrix(int rows, int cols) {
Matrix random = new Matrix(rows, cols);
return random.map(
new Matrix.Function() {
@Override
public double function(double x) {
return Math.random();
}
});
}
private Matrix applyFunctionDerivative(Matrix input) {
Matrix activated = (Matrix) input.clone();
if (function instanceof Softmax)
activated =
activated.map(
new Matrix.Function() {
@Override
public double function(double x) {
return Math.exp(x);
}
});
else
activated =
activated.map(
new Matrix.Function() {
@Override
public double function(double x) {
return function.derivative(x);
}
});
if (function instanceof Softmax) {
double sum = activated.sum();
if (sum != 0) activated = activated.multiply(1 / sum);
activated = activated.subtract(input);
activated =
activated.map(
new Matrix.Function() {
@Override
public double function(double x) {
return function.activate(x);
}
});
}
return activated;
}
/**
* Calculate the cross entropy error of the neural network.
*
* @param x The input to the neural network.
* @param y The expected output.
* @return The cross entropy error.
*/
public double crossEntropyError(Matrix x, Matrix y) {
Matrix y_ = predict(x);
double j =
y_.multiply(
y.map(
new Matrix.Function() {
@Override
public double function(double x) {
return Math.log(x);
}
}))
.sum();
return -j;
}
